Skip to content

M
models

PaddlePaddle / models
大约 1 年 前同步成功

通知 222
Star 6828
Fork 2962
M
models
提交 dc1b032d 编写于 作者: R ruri 提交者： qingqing01
浏览文件
操作

Refine Image Classification Codes&Docs (#1943) 

* Move models to legacy, rename models_name to models, refine README
* Refine eval/infer/train code
* Release ResNet 18/34, GoogleNet and ShuffleNet v2.
* Add RMSProp optimizer.
上级 b610d165
显示空白变更内容
内联 并排
Showing with 1658 addition and 1122 deletion
fluid/PaddleCV/image_classification/README.md
浏览文件 @ dc1b032d
......@@ -37,8 +37,6 @@ In the shell script ```download_imagenet2012.sh```, there are three steps to pr
train/n02483708/n02483708_2436.jpeg 369
train/n03998194/n03998194_7015.jpeg 741
train/n04523525/n04523525_38118.jpeg 884
train/n04596742/n04596742_3032.jpeg 909
train/n03208938/n03208938_7065.jpeg 535
...
```
* *val_list.txt*: label file of imagenet-2012 validation set, with each line seperated by ```SPACE```, like.
......@@ -46,8 +44,6 @@ train/n03208938/n03208938_7065.jpeg 535
val/ILSVRC2012_val_00000001.jpeg 65
val/ILSVRC2012_val_00000002.jpeg 970
val/ILSVRC2012_val_00000003.jpeg 230
val/ILSVRC2012_val_00000004.jpeg 809
val/ILSVRC2012_val_00000005.jpeg 516
...
```
......@@ -84,15 +80,14 @@ python train.py \
* **pretrained_model**: model path for pretraining. Default: None.
* **checkpoint**: the checkpoint path to resume. Default: None.
* **data_dir**: the data path. Default: "./data/ILSVRC2012".
* **model_category**: the category of models, ("models"|"models_name"). Default: "models_name".
* **fp16**: whether to enable half precisioin training with fp16. Default: False.
* **fp16**: whether to enable half precision training with fp16. Default: False.
* **scale_loss**: scale loss for fp16. Default: 1.0.
* **l2_decay**: L2_decay parameter. Default: 1e-4.
* **momentum_rate**: momentum_rate. Default: 0.9.
Or can start the training step by running the ```run.sh```.
**data reader introduction:** Data reader is defined in ```reader.py``` and ```reader_cv2.py```, Using CV2 reader can improve the speed of reading. In [training stage](#training-a-model), random crop and flipping are used, while center crop is used in [evaluation](#inference) and [inference](#inference) stages. Supported data augmentation includes:
**data reader introduction:** Data reader is defined in ```reader.py```and```reader_cv2.py```, Using CV2 reader can improve the speed of reading. In [training stage](#training-a-model-with-flexible-parameters), random crop and flipping are used, while center crop is used in [Evaluation](#evaluation) and [Inference](#inference) stages. Supported data augmentation includes:
* rotation
* color jitter
* random crop
......@@ -100,32 +95,11 @@ Or can start the training step by running the ```run.sh```.
* resize
* flipping
**training curve:** The training curve can be drawn based on training log. For example, the log from training AlexNet is like:
```
End pass 1, train_loss 6.23153877258, train_acc1 0.0150696625933, train_acc5 0.0552518665791, test_loss 5.41981744766, test_acc1 0.0519132651389, test_acc5 0.156150355935
End pass 2, train_loss 5.15442800522, train_acc1 0.0784279331565, train_acc5 0.211050540209, test_loss 4.45795249939, test_acc1 0.140469551086, test_acc5 0.333163291216
End pass 3, train_loss 4.51505613327, train_acc1 0.145300447941, train_acc5 0.331567406654, test_loss 3.86548018456, test_acc1 0.219443559647, test_acc5 0.446448504925
End pass 4, train_loss 4.12735557556, train_acc1 0.19437250495, train_acc5 0.405713528395, test_loss 3.56990146637, test_acc1 0.264536827803, test_acc5 0.507190704346
End pass 5, train_loss 3.87505435944, train_acc1 0.229518383741, train_acc5 0.453582793474, test_loss 3.35345435143, test_acc1 0.297349333763, test_acc5 0.54753267765
End pass 6, train_loss 3.6929500103, train_acc1 0.255628824234, train_acc5 0.487188398838, test_loss 3.17112898827, test_acc1 0.326953113079, test_acc5 0.581780135632
End pass 7, train_loss 3.55882954597, train_acc1 0.275381118059, train_acc5 0.511990904808, test_loss 3.03736782074, test_acc1 0.349035382271, test_acc5 0.606293857098
End pass 8, train_loss 3.45595097542, train_acc1 0.291462600231, train_acc5 0.530815005302, test_loss 2.96034455299, test_acc1 0.362228929996, test_acc5 0.617390751839
End pass 9, train_loss 3.3745200634, train_acc1 0.303871691227, train_acc5 0.545210540295, test_loss 2.93932366371, test_acc1 0.37129303813, test_acc5 0.623573005199
...
```
The error rate curves of AlexNet, ResNet50 and SE-ResNeXt-50 are shown in the figure below.
<p align="center">
<img src="images/curve.jpg" height=480 width=640 hspace='10'/> <br />
Training and validation Curves
</p>
## Using Mixed-Precision Training
You may add `--fp16 1` to start train using mixed precisioin training, which the training process will use float16 and the output model ("master" parameters) is saved as float32. You also may need to pass `--scale_loss` to overcome accuracy issues, usually `--scale_loss 8.0` will do.
You may add `--fp16=1` to start train using mixed precisioin training, which the training process will use float16 and the output model ("master" parameters) is saved as float32. You also may need to pass `--scale_loss` to overcome accuracy issues, usually `--scale_loss=8.0` will do.
Note that currently `--fp16` can not use together with `--with_mem_opt`, so pass `--with_mem_opt 0` to disable memory optimization pass.
Note that currently `--fp16` can not use together with `--with_mem_opt`, so pass `--with_mem_opt=0` to disable memory optimization pass.
## Finetuning
......@@ -156,20 +130,6 @@ python eval.py \
--pretrained_model=${path_to_pretrain_model}
```
According to the congfiguration of evaluation, the output log is like:
```
Testbatch 0,loss 2.1786134243, acc1 0.625,acc5 0.8125,time 0.48 sec
Testbatch 10,loss 0.898496925831, acc1 0.75,acc5 0.9375,time 0.51 sec
Testbatch 20,loss 1.32524681091, acc1 0.6875,acc5 0.9375,time 0.37 sec
Testbatch 30,loss 1.46830511093, acc1 0.5,acc5 0.9375,time 0.51 sec
Testbatch 40,loss 1.12802267075, acc1 0.625,acc5 0.9375,time 0.35 sec
Testbatch 50,loss 0.881597697735, acc1 0.8125,acc5 1.0,time 0.32 sec
Testbatch 60,loss 0.300163716078, acc1 0.875,acc5 1.0,time 0.48 sec
Testbatch 70,loss 0.692037761211, acc1 0.875,acc5 1.0,time 0.35 sec
Testbatch 80,loss 0.0969972759485, acc1 1.0,acc5 1.0,time 0.41 sec
...
```
## Inference
Inference is used to get prediction score or image features based on trained models.
```
......@@ -180,30 +140,10 @@ python infer.py \
--with_mem_opt=True \
--pretrained_model=${path_to_pretrain_model}
```
The output contains predication results, including maximum score (before softmax) and corresponding predicted label.
```
Test-0-score: [13.168352], class [491]
Test-1-score: [7.913302], class [975]
Test-2-score: [16.959702], class [21]
Test-3-score: [14.197695], class [383]
Test-4-score: [12.607652], class [878]
Test-5-score: [17.725458], class [15]
Test-6-score: [12.678599], class [118]
Test-7-score: [12.353498], class [505]
Test-8-score: [20.828007], class [747]
Test-9-score: [15.135801], class [315]
Test-10-score: [14.585114], class [920]
Test-11-score: [13.739927], class [679]
Test-12-score: [15.040644], class [386]
...
```
## Supported models and performances
Models consists of two categories: Models with specified parameters names in model definition and Models without specified parameters, Generate named model by indicating ```model_category = models_name```.
Models are trained by starting with learning rate ```0.1``` and decaying it by ```0.1``` after each pre-defined epoches, if not special introduced. Available top-1/top-5 validation accuracy on ImageNet 2012 are listed in table. Pretrained models can be downloaded by clicking related model names.
Available top-1/top-5 validation accuracy on ImageNet 2012 are listed in table. Pretrained models can be downloaded by clicking related model names.
- Released models: specify parameter names
......@@ -216,20 +156,12 @@ Models are trained by starting with learning rate ```0.1``` and decaying it by `
|[VGG19](https://paddle-imagenet-models-name.bj.bcebos.com/VGG19_pretrained.zip) | 72.56%/90.83% | 72.32%/90.98% |
|[MobileNetV1](http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.zip) | 70.91%/89.54% | 70.51%/89.35% |
|[MobileNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.zip) | 71.90%/90.55% | 71.53%/90.41% |
|[ResNet18](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar) | 70.85%/89.89% | 70.65%/89.89% |
|[ResNet34](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar) | 74.41%/92.03% | 74.13%/91.97% |
|[ResNet50](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.zip) | 76.35%/92.80% | 76.22%/92.92% |
|[ResNet101](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_pretrained.zip) | 77.49%/93.57% | 77.56%/93.64% |
|[ResNet152](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_pretrained.zip) | 78.12%/93.93% | 77.92%/93.87% |
|[SE_ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNext50_32x4d_pretrained.zip) | 78.50%/94.01% | 78.44%/93.96% |
|[SE_ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt101_32x4d_pretrained.zip) | 79.26%/94.22% | 79.12%/94.20% |
- Released models: not specify parameter names
**NOTE: These are trained by using model_category=models**
|model | top-1/top-5 accuracy(PIL)| top-1/top-5 accuracy(CV2) |
|- |:-: |:-:|
|[ResNet152](http://paddle-imagenet-models.bj.bcebos.com/ResNet152_pretrained.zip) | 78.18%/93.93% | 78.11%/94.04% |
|[SE_ResNeXt50_32x4d](http://paddle-imagenet-models.bj.bcebos.com/se_resnext_50_model.tar) | 78.32%/93.96% | 77.58%/93.73% |
|[GoogleNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogleNet_pretrained.tar) | 70.50%/89.59% | 70.27%/89.58% |
|[ShuffleNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNet_pretrained.tar) | | 69.48%/88.99% |
fluid/PaddleCV/image_classification/README_cn.md
浏览文件 @ dc1b032d
# 图像分类以及模型库
图像分类是计算机视觉的重要领域，它的目标是将图像分类到预定义的标签。近期，需要研究者提出很多不同种类的神经网络，并且极大的提升了分类算法的性能。本页将介绍如何使用PaddlePaddle进行图像分类，包括[数据准备](#data-preparation)、 [训练](#training-a-model)、[参数微调](#finetuning)、[模型评估](#evaluation)以及[模型推断](#inference)。
图像分类是计算机视觉的重要领域，它的目标是将图像分类到预定义的标签。近期，许多研究者提出很多不同种类的神经网络，并且极大的提升了分类算法的性能。本页将介绍如何使用PaddlePaddle进行图像分类。
---
## 内容
- [安装](#installation)
- [数据准备](#data-preparation)
- [模型训练](#training-a-model)
- [参数微调](#finetuning)
- [模型评估](#evaluation)
- [模型推断](#inference)
- [已有模型及其性能](#supported-models)
- [安装](#安装)
- [数据准备](#数据准备)
- [模型训练](#模型训练)
- [混合精度训练](#混合精度训练)
- [参数微调](#参数微调)
- [模型评估](#模型评估)
- [模型预测](#模型预测)
- [已有模型及其性能](#已有模型及其性能)
## 安装
在当前目录下运行样例代码需要PadddlePaddle Fluid的v0.13.0或以上的版本。如果你的运行环境中的PaddlePaddle低于此版本，请根据安装文档中的说明来更新PaddlePaddle。
在当前目录下运行样例代码需要PadddlePaddle Fluid的v0.13.0或以上的版本。如果你的运行环境中的PaddlePaddle低于此版本，请根据 [installation document](http://paddlepaddle.org/documentation/docs/zh/1.3/beginners_guide/install/index_cn.html) 中的说明来更新PaddlePaddle。
## 数据准备
......@@ -36,8 +37,6 @@ sh download_imagenet2012.sh
train/n02483708/n02483708_2436.jpeg 369
train/n03998194/n03998194_7015.jpeg 741
train/n04523525/n04523525_38118.jpeg 884
train/n04596742/n04596742_3032.jpeg 909
train/n03208938/n03208938_7065.jpeg 535
...
```
* *val_list.txt*: ImageNet-2012验证集合的标签文件，每一行采用"空格"分隔图像路径与标注，例如：
......@@ -45,10 +44,9 @@ train/n03208938/n03208938_7065.jpeg 535
val/ILSVRC2012_val_00000001.jpeg 65
val/ILSVRC2012_val_00000002.jpeg 970
val/ILSVRC2012_val_00000003.jpeg 230
val/ILSVRC2012_val_00000004.jpeg 809
val/ILSVRC2012_val_00000005.jpeg 516
...
```
注意：需要根据本地环境调整reader.py相关路径来正确读取数据。
## 模型训练
......@@ -66,22 +64,27 @@ python train.py \
--lr=0.1
```
**参数说明：**
* **model**: name model to use. Default: "SE_ResNeXt50_32x4d".
* **num_epochs**: the number of epochs. Default: 120.
* **batch_size**: the size of each mini-batch. Default: 256.
* **use_gpu**: whether to use GPU or not. Default: True.
* **total_images**: total number of images in the training set. Default: 1281167.
* **class_dim**: the class number of the classification task. Default: 1000.
* **image_shape**: input size of the network. Default: "3,224,224".
* **model_save_dir**: the directory to save trained model. Default: "output".
* **with_mem_opt**: whether to use memory optimization or not. Default: False.
* **lr_strategy**: learning rate changing strategy. Default: "piecewise_decay".
* **lr**: initialized learning rate. Default: 0.1.
* **pretrained_model**: model path for pretraining. Default: None.
* **checkpoint**: the checkpoint path to resume. Default: None.
* **model_category**: the category of models, ("models"|"models_name"). Default:"models_name".
**数据读取器说明：** 数据读取器定义在```reader.py```和```reader_cv2.py```中, 一般, CV2 reader可以提高数据读取速度, reader(PIL)可以得到相对更高的精度, 在[训练阶段](#training-a-model), 默认采用的增广方式是随机裁剪与水平翻转, 而在[评估](#inference)与[推断](#inference)阶段用的默认方式是中心裁剪。当前支持的数据增广方式有：
* **model**: 模型名称， 默认值: "SE_ResNeXt50_32x4d"
* **num_epochs**: 训练回合数，默认值: 120
* **batch_size**: 批大小，默认值: 256
* **use_gpu**: 是否在GPU上运行，默认值: True
* **total_images**: 图片数，ImageNet2012默认值: 1281167.
* **class_dim**: 类别数，默认值: 1000
* **image_shape**: 图片大小，默认值: "3,224,224"
* **model_save_dir**: 模型存储路径，默认值: "output/"
* **with_mem_opt**: 是否开启显存优化，默认值: False
* **lr_strategy**: 学习率变化策略，默认值: "piecewise_decay"
* **lr**: 初始学习率，默认值: 0.1
* **pretrained_model**: 预训练模型路径，默认值: None
* **checkpoint**: 用于继续训练的检查点（指定具体模型存储路径，如"output/SE_ResNeXt50_32x4d/100/"），默认值: None
* **fp16**: 是否开启混合精度训练，默认值: False
* **scale_loss**: 调整混合训练的loss scale值，默认值: 1.0
* **l2_decay**: l2_decay值，默认值: 1e-4
* **momentum_rate**: momentum_rate值，默认值: 0.9
在```run.sh```中有用于训练的脚本.
**数据读取器说明：** 数据读取器定义在```reader.py```和```reader_cv2.py```中。一般, CV2可以提高数据读取速度, PIL reader可以得到相对更高的精度, 我们现在默认基于PIL的数据读取器, 在[训练阶段](#模型训练), 默认采用的增广方式是随机裁剪与水平翻转, 而在[模型评估](#模型评估)与[模型预测](#模型预测)阶段用的默认方式是中心裁剪。当前支持的数据增广方式有：
* 旋转
* 颜色抖动
* 随机裁剪
......@@ -89,31 +92,11 @@ python train.py \
* 长宽调整
* 水平翻转
**训练曲线：** 通过训练过程中的日志可以画出训练曲线。举个例子，训练AlexNet出来的日志如下所示：
```
End pass 1, train_loss 6.23153877258, train_acc1 0.0150696625933, train_acc5 0.0552518665791, test_loss 5.41981744766, test_acc1 0.0519132651389, test_acc5 0.156150355935
End pass 2, train_loss 5.15442800522, train_acc1 0.0784279331565, train_acc5 0.211050540209, test_loss 4.45795249939, test_acc1 0.140469551086, test_acc5 0.333163291216
End pass 3, train_loss 4.51505613327, train_acc1 0.145300447941, train_acc5 0.331567406654, test_loss 3.86548018456, test_acc1 0.219443559647, test_acc5 0.446448504925
End pass 4, train_loss 4.12735557556, train_acc1 0.19437250495, train_acc5 0.405713528395, test_loss 3.56990146637, test_acc1 0.264536827803, test_acc5 0.507190704346
End pass 5, train_loss 3.87505435944, train_acc1 0.229518383741, train_acc5 0.453582793474, test_loss 3.35345435143, test_acc1 0.297349333763, test_acc5 0.54753267765
End pass 6, train_loss 3.6929500103, train_acc1 0.255628824234, train_acc5 0.487188398838, test_loss 3.17112898827, test_acc1 0.326953113079, test_acc5 0.581780135632
End pass 7, train_loss 3.55882954597, train_acc1 0.275381118059, train_acc5 0.511990904808, test_loss 3.03736782074, test_acc1 0.349035382271, test_acc5 0.606293857098
End pass 8, train_loss 3.45595097542, train_acc1 0.291462600231, train_acc5 0.530815005302, test_loss 2.96034455299, test_acc1 0.362228929996, test_acc5 0.617390751839
End pass 9, train_loss 3.3745200634, train_acc1 0.303871691227, train_acc5 0.545210540295, test_loss 2.93932366371, test_acc1 0.37129303813, test_acc5 0.623573005199
...
```
下图给出了AlexNet、ResNet50以及SE-ResNeXt-50网络的错误率曲线：
<p align="center">
<img src="images/curve.jpg" height=480 width=640 hspace='10'/> <br />
训练集合与验证集合上的错误率曲线
</p>
## 混合精度训练
可以通过开启`--fp16 1`启动混合精度训练，这样训练过程会使用float16数据，并输出float32的模型参数（"master"参数）。您可能需要同时传入`--scale_loss`来解决fp16训练的精度问题，通常传入`--scale_loss 8.0`即可。
可以通过开启`--fp16=True`启动混合精度训练，这样训练过程会使用float16数据，并输出float32的模型参数（"master"参数）。您可能需要同时传入`--scale_loss`来解决fp16训练的精度问题，通常传入`--scale_loss=8.0`即可。
注意，目前混合精度训练不能和内存优化功能同时使用，所以需要传`--with_mem_opt 0`这个参数来禁用内存优化功能。
注意，目前混合精度训练不能和内存优化功能同时使用，所以需要传`--with_mem_opt=False`这个参数来禁用内存优化功能。
## 参数微调
......@@ -133,7 +116,7 @@ python train.py
```
## 模型评估
模型评估是指对训练完毕的模型评估各类性能指标。用户可以下载[预训练模型](#supported-models)并且设置```path_to_pretrain_model```为模型所在路径。运行如下的命令，可以获得一个模型top-1/top-5精度:
模型评估是指对训练完毕的模型评估各类性能指标。用户可以下载[已有模型及其性能](#已有模型及其性能)并且设置```path_to_pretrain_model```为模型所在路径。运行如下的命令，可以获得一个模型top-1/top-5精度:
```
python eval.py \
--model=SE_ResNeXt50_32x4d \
......@@ -144,23 +127,8 @@ python eval.py \
--pretrained_model=${path_to_pretrain_model}
```
根据这个评估程序的配置，输出日志形式如下：
```
Testbatch 0,loss 2.1786134243, acc1 0.625,acc5 0.8125,time 0.48 sec
Testbatch 10,loss 0.898496925831, acc1 0.75,acc5 0.9375,time 0.51 sec
Testbatch 20,loss 1.32524681091, acc1 0.6875,acc5 0.9375,time 0.37 sec
Testbatch 30,loss 1.46830511093, acc1 0.5,acc5 0.9375,time 0.51 sec
Testbatch 40,loss 1.12802267075, acc1 0.625,acc5 0.9375,time 0.35 sec
Testbatch 50,loss 0.881597697735, acc1 0.8125,acc5 1.0,time 0.32 sec
Testbatch 60,loss 0.300163716078, acc1 0.875,acc5 1.0,time 0.48 sec
Testbatch 70,loss 0.692037761211, acc1 0.875,acc5 1.0,time 0.35 sec
Testbatch 80,loss 0.0969972759485, acc1 1.0,acc5 1.0,time 0.41 sec
...
```
## 模型推断
模型推断可以获取一个模型的预测分数或者图像的特征：
## 模型预测
模型预测可以获取一个模型的预测分数或者图像的特征：
```
python infer.py \
--model=SE_ResNeXt50_32x4d \
......@@ -169,31 +137,12 @@ python infer.py \
--with_mem_opt=True \
--pretrained_model=${path_to_pretrain_model}
```
输出的预测结果包括最高分数(未经过softmax处理)以及相应的预测标签。
```
Test-0-score: [13.168352], class [491]
Test-1-score: [7.913302], class [975]
Test-2-score: [16.959702], class [21]
Test-3-score: [14.197695], class [383]
Test-4-score: [12.607652], class [878]
Test-5-score: [17.725458], class [15]
Test-6-score: [12.678599], class [118]
Test-7-score: [12.353498], class [505]
Test-8-score: [20.828007], class [747]
Test-9-score: [15.135801], class [315]
Test-10-score: [14.585114], class [920]
Test-11-score: [13.739927], class [679]
Test-12-score: [15.040644], class [386]
...
```
## 已有模型及其性能
Models包括两种模型：带有参数名字的模型，和不带有参数名字的模型。通过设置 ```model_category = models_name```来训练带有参数名字的模型。
表格中列出了在"models"目录下支持的神经网络种类，并且给出了已完成训练的模型在ImageNet-2012验证集合上的top-1/top-5精度；如无特征说明，训练模型的初始学习率为```0.1```，每隔预定的epochs会下降```0.1```。预训练模型可以通过点击相应模型的名称进行下载。
表格中列出了在```models```目录下支持的图像分类模型，并且给出了已完成训练的模型在ImageNet-2012验证集合上的top-1/top-5精度，
可以通过点击相应模型的名称下载相应预训练模型。
- Released models: specify parameter names
- Released models:
|model | top-1/top-5 accuracy(PIL)| top-1/top-5 accuracy(CV2) |
|- |:-: |:-:|
......@@ -204,17 +153,12 @@ Models包括两种模型：带有参数名字的模型，和不带有参数名
|[VGG19](https://paddle-imagenet-models-name.bj.bcebos.com/VGG19_pretrained.zip) | 72.56%/90.83% | 72.32%/90.98% |
|[MobileNetV1](http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.zip) | 70.91%/89.54% | 70.51%/89.35% |
|[MobileNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.zip) | 71.90%/90.55% | 71.53%/90.41% |
|[ResNet18](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet18_pretrained.tar) | 70.85%/89.89% | 70.65%/89.89% |
|[ResNet34](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar) | 74.41%/92.03% | 74.13%/91.97% |
|[ResNet50](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.zip) | 76.35%/92.80% | 76.22%/92.92% |
|[ResNet101](http://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_pretrained.zip) | 77.49%/93.57% | 77.56%/93.64% |
|[ResNet152](https://paddle-imagenet-models-name.bj.bcebos.com/ResNet152_pretrained.zip) | 78.12%/93.93% | 77.92%/93.87% |
|[SE_ResNeXt50_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNext50_32x4d_pretrained.zip) | 78.50%/94.01% | 78.44%/93.96% |
|[SE_ResNeXt101_32x4d](https://paddle-imagenet-models-name.bj.bcebos.com/SE_ResNeXt101_32x4d_pretrained.zip) | 79.26%/94.22% | 79.12%/94.20% |
- Released models: not specify parameter names
**注意：这是model_category = models 的预训练模型**
|model | top-1/top-5 accuracy(PIL)| top-1/top-5 accuracy(CV2) |
|- |:-: |:-:|
|[ResNet152](http://paddle-imagenet-models.bj.bcebos.com/ResNet152_pretrained.zip) | 78.18%/93.93% | 78.11%/94.04% |
|[SE_ResNeXt50_32x4d](http://paddle-imagenet-models.bj.bcebos.com/se_resnext_50_model.tar) | 78.32%/93.96% | 77.58%/93.73% |
|[GoogleNet](https://paddle-imagenet-models-name.bj.bcebos.com/GoogleNet_pretrained.tar) | 70.50%/89.59% | 70.27%/89.58% |
|[ShuffleNetV2](https://paddle-imagenet-models-name.bj.bcebos.com/ShuffleNet_pretrained.tar) | | 69.48%/88.99% |
fluid/PaddleCV/image_classification/eval.py
浏览文件 @ dc1b032d
......@@ -7,12 +7,12 @@ import time
import sys
import paddle
import paddle.fluid as fluid
#import reader_cv2 as reader
import reader as reader
import argparse
import functools
import models
from utils.learning_rate import cosine_decay
from utility import add_arguments, print_arguments
from utils.utility import add_arguments, print_arguments
import math
parser = argparse.ArgumentParser(description=__doc__)
......@@ -25,22 +25,9 @@ add_arg('image_shape', str, "3,224,224", "Input image size")
add_arg('with_mem_opt', bool, True, "Whether to use memory optimization or not.")
add_arg('pretrained_model', str, None, "Whether to use pretrained model.")
add_arg('model', str, "SE_ResNeXt50_32x4d", "Set the network to use.")
add_arg('model_category', str, "models_name", "Whether to use models_name or not, valid value:'models','models_name'." )
# yapf: enable
def set_models(model_category):
global models
assert model_category in ["models", "models_name"
], "{} is not in lists: {}".format(
model_category, ["models", "models_name"])
if model_category == "models_name":
import models_name as models
else:
import models as models
def eval(args):
# parameters from arguments
class_dim = args.class_dim
......@@ -119,7 +106,7 @@ def eval(args):
if batch_id % 10 == 0:
print("Testbatch {0},loss {1}, "
"acc1 {2},acc5 {3},time {4}".format(batch_id, \
loss, acc1, acc5, \
"%.5f"%loss,"%.5f"%acc1, "%.5f"%acc5, \
"%2.2f sec" % period))
sys.stdout.flush()
......@@ -128,14 +115,13 @@ def eval(args):
test_acc5 = np.sum(test_info[2]) / cnt
print("Test_loss {0}, test_acc1 {1}, test_acc5 {2}".format(
test_loss, test_acc1, test_acc5))
"%.5f"%test_loss, "%.5f"%test_acc1, "%.5f"%test_acc5))
sys.stdout.flush()
def main():
args = parser.parse_args()
print_arguments(args)
set_models(args.model_category)
eval(args)
......
fluid/PaddleCV/image_classification/infer.py
浏览文件 @ dc1b032d
......@@ -10,7 +10,9 @@ import paddle.fluid as fluid
import reader
import argparse
import functools
from utility import add_arguments, print_arguments
import models
import utils
from utils.utility import add_arguments,print_arguments
import math
parser = argparse.ArgumentParser(description=__doc__)
......@@ -22,25 +24,14 @@ add_arg('image_shape', str, "3,224,224", "Input image size")
add_arg('with_mem_opt', bool, True, "Whether to use memory optimization or not.")
add_arg('pretrained_model', str, None, "Whether to use pretrained model.")
add_arg('model', str, "SE_ResNeXt50_32x4d", "Set the network to use.")
add_arg('model_category', str, "models_name", "Whether to use models_name or not, valid value:'models','models_name'." )
add_arg('save_inference', bool, False, "Whether to save inference model or not")
# yapf: enable
def set_models(model_category):
global models
assert model_category in ["models", "models_name"
], "{} is not in lists: {}".format(
model_category, ["models", "models_name"])
if model_category == "models_name":
import models_name as models
else:
import models as models
def infer(args):
# parameters from arguments
class_dim = args.class_dim
model_name = args.model
save_inference = args.save_inference
pretrained_model = args.pretrained_model
with_memory_optimization = args.with_mem_opt
image_shape = [int(m) for m in args.image_shape.split(",")]
......@@ -52,7 +43,7 @@ def infer(args):
# model definition
model = models.__dict__[model_name]()
if model_name is "GoogleNet":
if model_name == "GoogleNet":
out, _, _ = model.net(input=image, class_dim=class_dim)
else:
out = model.net(input=image, class_dim=class_dim)
......@@ -60,7 +51,7 @@ def infer(args):
test_program = fluid.default_main_program().clone(for_test=True)
fetch_list = [out.name]
if with_memory_optimization:
if with_memory_optimization and not save_inference:
fluid.memory_optimize(
fluid.default_main_program(), skip_opt_set=set(fetch_list))
......@@ -74,7 +65,17 @@ def infer(args):
return os.path.exists(os.path.join(pretrained_model, var.name))
fluid.io.load_vars(exe, pretrained_model, predicate=if_exist)
if save_inference:
fluid.io.save_inference_model(
dirname=model_name,
feeded_var_names=['image'],
main_program=test_program,
target_vars=out,
executor=exe,
model_filename='model',
params_filename='params')
print("model: ",model_name," is already saved")
exit(0)
test_batch_size = 1
test_reader = paddle.batch(reader.test(), batch_size=test_batch_size)
feeder = fluid.DataFeeder(place=place, feed_list=[image])
......@@ -94,7 +95,6 @@ def infer(args):
def main():
args = parser.parse_args()
print_arguments(args)
set_models(args.model_category)
infer(args)
......
fluid/PaddleCV/image_classification/legacy/README.md 0 → 100644
浏览文件 @ dc1b032d
For historical reasons, We keep "no name" models here, which are different from "specified name" models.
**NOTE: Training the models in legacy folder will generate models without specified parameters.**
- **Released models: not specify parameter names**
|model | top-1/top-5 accuracy(PIL)| top-1/top-5 accuracy(CV2) |
|- |:-: |:-:|
|[ResNet152](http://paddle-imagenet-models.bj.bcebos.com/ResNet152_pretrained.zip) | 78.18%/93.93% | 78.11%/94.04% |
|[SE_ResNeXt50_32x4d](http://paddle-imagenet-models.bj.bcebos.com/se_resnext_50_model.tar) | 78.32%/93.96% | 77.58%/93.73% |
fluid/PaddleCV/image_classification/models_name/__init__.py → fluid/PaddleCV/image_classification/legacy/models/__init__.py
浏览文件 @ dc1b032d
......@@ -4,7 +4,9 @@ from .mobilenet_v2 import MobileNetV2
from .googlenet import GoogleNet
from .vgg import VGG11, VGG13, VGG16, VGG19
from .resnet import ResNet50, ResNet101, ResNet152
from .resnet_dist import DistResNet
from .inception_v4 import InceptionV4
from .se_resnext import SE_ResNeXt50_32x4d, SE_ResNeXt101_32x4d, SE_ResNeXt152_32x4d
from .dpn import DPN68, DPN92, DPN98, DPN107, DPN131
from .shufflenet_v2 import ShuffleNetV2_x0_5, ShuffleNetV2_x1_0, ShuffleNetV2_x1_5, ShuffleNetV2_x2_0
from .fast_imagenet import FastImageNet
fluid/PaddleCV/image_classification/models_name/alexnet.py → fluid/PaddleCV/image_classification/legacy/models/alexnet.py
浏览文件 @ dc1b032d
......@@ -26,9 +26,6 @@ class AlexNet():
def net(self, input, class_dim=1000):
stdv = 1.0 / math.sqrt(input.shape[1] * 11 * 11)
layer_name = [
"conv1", "conv2", "conv3", "conv4", "conv5", "fc6", "fc7", "fc8"
]
conv1 = fluid.layers.conv2d(
input=input,
num_filters=64,
......@@ -38,11 +35,9 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[0] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[0] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
pool1 = fluid.layers.pool2d(
input=conv1,
pool_size=3,
......@@ -60,11 +55,9 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[1] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[1] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
pool2 = fluid.layers.pool2d(
input=conv2,
pool_size=3,
......@@ -82,11 +75,9 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[2] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[2] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
stdv = 1.0 / math.sqrt(conv3.shape[1] * 3 * 3)
conv4 = fluid.layers.conv2d(
......@@ -98,11 +89,9 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[3] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[3] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
stdv = 1.0 / math.sqrt(conv4.shape[1] * 3 * 3)
conv5 = fluid.layers.conv2d(
......@@ -114,11 +103,9 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[4] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[4] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
pool5 = fluid.layers.pool2d(
input=conv5,
pool_size=3,
......@@ -127,42 +114,36 @@ class AlexNet():
pool_type='max')
drop6 = fluid.layers.dropout(x=pool5, dropout_prob=0.5)
stdv = 1.0 / math.sqrt(drop6.shape[1] * drop6.shape[2] *
drop6.shape[3] * 1.0)
fc6 = fluid.layers.fc(
input=drop6,
size=4096,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[5] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[5] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
drop7 = fluid.layers.dropout(x=fc6, dropout_prob=0.5)
stdv = 1.0 / math.sqrt(drop7.shape[1] * 1.0)
stdv = 1.0 / math.sqrt(drop7.shape[1] * 1.0)
fc7 = fluid.layers.fc(
input=drop7,
size=4096,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[6] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[6] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
stdv = 1.0 / math.sqrt(fc7.shape[1] * 1.0)
out = fluid.layers.fc(
input=fc7,
size=class_dim,
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[7] + "_offset"),
initializer=fluid.initializer.Uniform(-stdv, stdv)),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[7] + "_weights"))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
return out
fluid/PaddleCV/image_classification/models_name/dpn.py → fluid/PaddleCV/image_classification/legacy/models/dpn.py
浏览文件 @ dc1b032d
......@@ -7,7 +7,6 @@ import time
import sys
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ["DPN", "DPN68", "DPN92", "DPN98", "DPN107", "DPN131"]
......@@ -53,30 +52,16 @@ class DPN(object):
padding=init_padding,
groups=1,
act=None,
bias_attr=False,
name="conv1",
param_attr=ParamAttr(name="conv1_weights"), )
bias_attr=False)
conv1_x_1 = fluid.layers.batch_norm(
input=conv1_x_1,
act='relu',
is_test=False,
name="conv1_bn",
param_attr=ParamAttr(name='conv1_bn_scale'),
bias_attr=ParamAttr('conv1_bn_offset'),
moving_mean_name='conv1_bn_mean',
moving_variance_name='conv1_bn_variance', )
input=conv1_x_1, act='relu', is_test=False)
convX_x_x = fluid.layers.pool2d(
input=conv1_x_1,
pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max',
name="pool1")
pool_type='max')
#conv2 - conv5
match_list, num = [], 0
for gc in range(4):
bw = bws[gc]
inc = inc_sec[gc]
......@@ -84,46 +69,32 @@ class DPN(object):
if gc == 0:
_type1 = 'proj'
_type2 = 'normal'
match = 1
else:
_type1 = 'down'
_type2 = 'normal'
match = match + k_sec[gc - 1]
match_list.append(match)
convX_x_x = self.dual_path_factory(
convX_x_x, R, R, bw, inc, G, _type1, name="dpn" + str(match))
convX_x_x = self.dual_path_factory(convX_x_x, R, R, bw, inc, G,
_type1)
for i_ly in range(2, k_sec[gc] + 1):
num += 1
if num in match_list:
num += 1
convX_x_x = self.dual_path_factory(
convX_x_x, R, R, bw, inc, G, _type2, name="dpn" + str(num))
convX_x_x = self.dual_path_factory(convX_x_x, R, R, bw, inc, G,
_type2)
conv5_x_x = fluid.layers.concat(convX_x_x, axis=1)
conv5_x_x = fluid.layers.batch_norm(
input=conv5_x_x,
act='relu',
is_test=False,
name="final_concat_bn",
param_attr=ParamAttr(name='final_concat_bn_scale'),
bias_attr=ParamAttr('final_concat_bn_offset'),
moving_mean_name='final_concat_bn_mean',
moving_variance_name='final_concat_bn_variance', )
input=conv5_x_x, act='relu', is_test=False)
pool5 = fluid.layers.pool2d(
input=conv5_x_x,
pool_size=7,
pool_stride=1,
pool_padding=0,
pool_type='avg', )
pool_type='avg')
#stdv = 1.0 / math.sqrt(pool5.shape[1] * 1.0)
stdv = 0.01
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
fc6 = fluid.layers.fc(input=pool5,
size=class_dim,
param_attr=param_attr,
name="fc6")
param_attr=param_attr)
return fc6
......@@ -201,8 +172,7 @@ class DPN(object):
num_1x1_c,
inc,
G,
_type='normal',
name=None):
_type='normal'):
kw = 3
kh = 3
pw = (kw - 1) // 2
......@@ -231,50 +201,35 @@ class DPN(object):
num_filter=(num_1x1_c + 2 * inc),
kernel=(1, 1),
pad=(0, 0),
stride=(key_stride, key_stride),
name=name + "_match")
stride=(key_stride, key_stride))
data_o1, data_o2 = fluid.layers.split(
c1x1_w,
num_or_sections=[num_1x1_c, 2 * inc],
dim=1,
name=name + "_match_conv_Slice")
c1x1_w, num_or_sections=[num_1x1_c, 2 * inc], dim=1)
else:
data_o1 = data[0]
data_o2 = data[1]
# MAIN
c1x1_a = self.bn_ac_conv(
data=data_in,
num_filter=num_1x1_a,
kernel=(1, 1),
pad=(0, 0),
name=name + "_conv1")
data=data_in, num_filter=num_1x1_a, kernel=(1, 1), pad=(0, 0))
c3x3_b = self.bn_ac_conv(
data=c1x1_a,
num_filter=num_3x3_b,
kernel=(kw, kh),
pad=(pw, ph),
stride=(key_stride, key_stride),
num_group=G,
name=name + "_conv2")
num_group=G)
c1x1_c = self.bn_ac_conv(
data=c3x3_b,
num_filter=(num_1x1_c + inc),
kernel=(1, 1),
pad=(0, 0),
name=name + "_conv3")
pad=(0, 0))
c1x1_c1, c1x1_c2 = fluid.layers.split(
c1x1_c,
num_or_sections=[num_1x1_c, inc],
dim=1,
name=name + "_conv3_Slice")
c1x1_c, num_or_sections=[num_1x1_c, inc], dim=1)
# OUTPUTS
summ = fluid.layers.elementwise_add(
x=data_o1, y=c1x1_c1, name=name + "_elewise")
dense = fluid.layers.concat(
[data_o2, c1x1_c2], axis=1, name=name + "_concat")
summ = fluid.layers.elementwise_add(x=data_o1, y=c1x1_c1)
dense = fluid.layers.concat([data_o2, c1x1_c2], axis=1)
return [summ, dense]
......@@ -284,17 +239,8 @@ class DPN(object):
kernel,
pad,
stride=(1, 1),
num_group=1,
name=None):
bn_ac = fluid.layers.batch_norm(
input=data,
act='relu',
is_test=False,
name=name + '.output.1',
param_attr=ParamAttr(name=name + '_bn_scale'),
bias_attr=ParamAttr(name + '_bn_offset'),
moving_mean_name=name + '_bn_mean',
moving_variance_name=name + '_bn_variance', )
num_group=1):
bn_ac = fluid.layers.batch_norm(input=data, act='relu', is_test=False)
bn_ac_conv = fluid.layers.conv2d(
input=bn_ac,
num_filters=num_filter,
......@@ -303,8 +249,7 @@ class DPN(object):
padding=pad,
groups=num_group,
act=None,
bias_attr=False,
param_attr=ParamAttr(name=name + "_weights"))
bias_attr=False)
return bn_ac_conv
......@@ -314,7 +259,7 @@ def DPN68():
def DPN92():
onvodel = DPN(layers=92)
model = DPN(layers=92)
return model
......
fluid/PaddleCV/image_classification/models_name/googlenet.py → fluid/PaddleCV/image_classification/legacy/models/googlenet.py
浏览文件 @ dc1b032d
......@@ -3,7 +3,6 @@ from __future__ import division
from __future__ import print_function
import paddle
import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
__all__ = ['GoogleNet']
......@@ -30,13 +29,11 @@ class GoogleNet():
filter_size,
stride=1,
groups=1,
act=None,
name=None):
act=None):
channels = input.shape[1]
stdv = (3.0 / (filter_size**2 * channels))**0.5
param_attr = ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + "_weights")
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -46,63 +43,43 @@ class GoogleNet():
groups=groups,
act=act,
param_attr=param_attr,
bias_attr=False,
name=name)
bias_attr=False)
return conv
def xavier(self, channels, filter_size, name):
def xavier(self, channels, filter_size):
stdv = (3.0 / (filter_size**2 * channels))**0.5
param_attr = ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + "_weights")
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
return param_attr
def inception(self,
input,
channels,
filter1,
filter3R,
filter3,
filter5R,
filter5,
proj,
name=None):
def inception(self, name, input, channels, filter1, filter3R, filter3,
filter5R, filter5, proj):
conv1 = self.conv_layer(
input=input,
num_filters=filter1,
filter_size=1,
stride=1,
act=None,
name="inception_" + name + "_1x1")
input=input, num_filters=filter1, filter_size=1, stride=1, act=None)
conv3r = self.conv_layer(
input=input,
num_filters=filter3R,
filter_size=1,
stride=1,
act=None,
name="inception_" + name + "_3x3_reduce")
act=None)
conv3 = self.conv_layer(
input=conv3r,
num_filters=filter3,
filter_size=3,
stride=1,
act=None,
name="inception_" + name + "_3x3")
act=None)
conv5r = self.conv_layer(
input=input,
num_filters=filter5R,
filter_size=1,
stride=1,
act=None,
name="inception_" + name + "_5x5_reduce")
act=None)
conv5 = self.conv_layer(
input=conv5r,
num_filters=filter5,
filter_size=5,
stride=1,
act=None,
name="inception_" + name + "_5x5")
act=None)
pool = fluid.layers.pool2d(
input=input,
pool_size=3,
......@@ -110,124 +87,81 @@ class GoogleNet():
pool_padding=1,
pool_type='max')
convprj = fluid.layers.conv2d(
input=pool,
filter_size=1,
num_filters=proj,
stride=1,
padding=0,
name="inception_" + name + "_3x3_proj",
param_attr=ParamAttr(
name="inception_" + name + "_3x3_proj_weights"),
bias_attr=False)
input=pool, filter_size=1, num_filters=proj, stride=1, padding=0)
cat = fluid.layers.concat(input=[conv1, conv3, conv5, convprj], axis=1)
cat = fluid.layers.relu(cat)
return cat
def net(self, input, class_dim=1000):
conv = self.conv_layer(
input=input,
num_filters=64,
filter_size=7,
stride=2,
act=None,
name="conv1")
input=input, num_filters=64, filter_size=7, stride=2, act=None)
pool = fluid.layers.pool2d(
input=conv, pool_size=3, pool_type='max', pool_stride=2)
conv = self.conv_layer(
input=pool,
num_filters=64,
filter_size=1,
stride=1,
act=None,
name="conv2_1x1")
input=pool, num_filters=64, filter_size=1, stride=1, act=None)
conv = self.conv_layer(
input=conv,
num_filters=192,
filter_size=3,
stride=1,
act=None,
name="conv2_3x3")
input=conv, num_filters=192, filter_size=3, stride=1, act=None)
pool = fluid.layers.pool2d(
input=conv, pool_size=3, pool_type='max', pool_stride=2)
ince3a = self.inception(pool, 192, 64, 96, 128, 16, 32, 32, "ince3a")
ince3b = self.inception(ince3a, 256, 128, 128, 192, 32, 96, 64,
"ince3b")
ince3a = self.inception("ince3a", pool, 192, 64, 96, 128, 16, 32, 32)
ince3b = self.inception("ince3b", ince3a, 256, 128, 128, 192, 32, 96,
64)
pool3 = fluid.layers.pool2d(
input=ince3b, pool_size=3, pool_type='max', pool_stride=2)
ince4a = self.inception(pool3, 480, 192, 96, 208, 16, 48, 64, "ince4a")
ince4b = self.inception(ince4a, 512, 160, 112, 224, 24, 64, 64,
"ince4b")
ince4c = self.inception(ince4b, 512, 128, 128, 256, 24, 64, 64,
"ince4c")
ince4d = self.inception(ince4c, 512, 112, 144, 288, 32, 64, 64,
"ince4d")
ince4e = self.inception(ince4d, 528, 256, 160, 320, 32, 128, 128,
"ince4e")
ince4a = self.inception("ince4a", pool3, 480, 192, 96, 208, 16, 48, 64)
ince4b = self.inception("ince4b", ince4a, 512, 160, 112, 224, 24, 64,
64)
ince4c = self.inception("ince4c", ince4b, 512, 128, 128, 256, 24, 64,
64)
ince4d = self.inception("ince4d", ince4c, 512, 112, 144, 288, 32, 64,
64)
ince4e = self.inception("ince4e", ince4d, 528, 256, 160, 320, 32, 128,
128)
pool4 = fluid.layers.pool2d(
input=ince4e, pool_size=3, pool_type='max', pool_stride=2)
ince5a = self.inception(pool4, 832, 256, 160, 320, 32, 128, 128,
"ince5a")
ince5b = self.inception(ince5a, 832, 384, 192, 384, 48, 128, 128,
"ince5b")
ince5a = self.inception("ince5a", pool4, 832, 256, 160, 320, 32, 128,
128)
ince5b = self.inception("ince5b", ince5a, 832, 384, 192, 384, 48, 128,
128)
pool5 = fluid.layers.pool2d(
input=ince5b, pool_size=7, pool_type='avg', pool_stride=7)
dropout = fluid.layers.dropout(x=pool5, dropout_prob=0.4)
out = fluid.layers.fc(input=dropout,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1, "out"),
name="out",
bias_attr=ParamAttr(name="out_offset"))
param_attr=self.xavier(1024, 1))
pool_o1 = fluid.layers.pool2d(
input=ince4a, pool_size=5, pool_type='avg', pool_stride=3)
conv_o1 = self.conv_layer(
input=pool_o1,
num_filters=128,
filter_size=1,
stride=1,
act=None,
name="conv_o1")
input=pool_o1, num_filters=128, filter_size=1, stride=1, act=None)
fc_o1 = fluid.layers.fc(input=conv_o1,
size=1024,
act='relu',
param_attr=self.xavier(2048, 1, "fc_o1"),
name="fc_o1",
bias_attr=ParamAttr(name="fc_o1_offset"))
param_attr=self.xavier(2048, 1))
dropout_o1 = fluid.layers.dropout(x=fc_o1, dropout_prob=0.7)
out1 = fluid.layers.fc(input=dropout_o1,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1, "out1"),
name="out1",
bias_attr=ParamAttr(name="out1_offset"))
param_attr=self.xavier(1024, 1))
pool_o2 = fluid.layers.pool2d(
input=ince4d, pool_size=5, pool_type='avg', pool_stride=3)
conv_o2 = self.conv_layer(
input=pool_o2,
num_filters=128,
filter_size=1,
stride=1,
act=None,
name="conv_o2")
input=pool_o2, num_filters=128, filter_size=1, stride=1, act=None)
fc_o2 = fluid.layers.fc(input=conv_o2,
size=1024,
act='relu',
param_attr=self.xavier(2048, 1, "fc_o2"),
name="fc_o2",
bias_attr=ParamAttr(name="fc_o2_offset"))
param_attr=self.xavier(2048, 1))
dropout_o2 = fluid.layers.dropout(x=fc_o2, dropout_prob=0.7)
out2 = fluid.layers.fc(input=dropout_o2,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1, "out2"),
name="out2",
bias_attr=ParamAttr(name="out2_offset"))
param_attr=self.xavier(1024, 1))
# last fc layer is "out"
return out, out1, out2
fluid/PaddleCV/image_classification/models_name/inception_v4.py → fluid/PaddleCV/image_classification/legacy/models/inception_v4.py
浏览文件 @ dc1b032d
......@@ -4,7 +4,6 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ['InceptionV4']
......@@ -29,15 +28,15 @@ class InceptionV4():
x = self.inception_stem(input)
for i in range(4):
x = self.inceptionA(x, name=str(i + 1))
x = self.inceptionA(x)
x = self.reductionA(x)
for i in range(7):
x = self.inceptionB(x, name=str(i + 1))
x = self.inceptionB(x)
x = self.reductionB(x)
for i in range(3):
x = self.inceptionC(x, name=str(i + 1))
x = self.inceptionC(x)
pool = fluid.layers.pool2d(
input=x, pool_size=8, pool_type='avg', global_pooling=True)
......@@ -48,12 +47,8 @@ class InceptionV4():
out = fluid.layers.fc(
input=drop,
size=class_dim,
param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name="final_fc_weights"),
bias_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name="final_fc_offset"))
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
return out
def conv_bn_layer(self,
......@@ -63,8 +58,7 @@ class InceptionV4():
stride=1,
padding=0,
groups=1,
act='relu',
name=None):
act='relu'):
conv = fluid.layers.conv2d(
input=data,
num_filters=num_filters,
......@@ -73,58 +67,32 @@ class InceptionV4():
padding=padding,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False,
name=name)
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
name=bn_name,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def inception_stem(self, data, name=None):
conv = self.conv_bn_layer(
data, 32, 3, stride=2, act='relu', name="conv1_3x3_s2")
conv = self.conv_bn_layer(conv, 32, 3, act='relu', name="conv2_3x3_s1")
conv = self.conv_bn_layer(
conv, 64, 3, padding=1, act='relu', name="conv3_3x3_s1")
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
def inception_stem(self, data):
conv = self.conv_bn_layer(data, 32, 3, stride=2, act='relu')
conv = self.conv_bn_layer(conv, 32, 3, act='relu')
conv = self.conv_bn_layer(conv, 64, 3, padding=1, act='relu')
pool1 = fluid.layers.pool2d(
input=conv, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(
conv, 96, 3, stride=2, act='relu', name="inception_stem1_3x3_s2")
conv2 = self.conv_bn_layer(conv, 96, 3, stride=2, act='relu')
concat = fluid.layers.concat([pool1, conv2], axis=1)
conv1 = self.conv_bn_layer(
concat, 64, 1, act='relu', name="inception_stem2_3x3_reduce")
conv1 = self.conv_bn_layer(
conv1, 96, 3, act='relu', name="inception_stem2_3x3")
conv1 = self.conv_bn_layer(concat, 64, 1, act='relu')
conv1 = self.conv_bn_layer(conv1, 96, 3, act='relu')
conv2 = self.conv_bn_layer(concat, 64, 1, act='relu')
conv2 = self.conv_bn_layer(
concat, 64, 1, act='relu', name="inception_stem2_1x7_reduce")
conv2 = self.conv_bn_layer(
conv2,
64, (7, 1),
padding=(3, 0),
act='relu',
name="inception_stem2_1x7")
conv2 = self.conv_bn_layer(
conv2,
64, (1, 7),
padding=(0, 3),
act='relu',
name="inception_stem2_7x1")
conv2, 64, (7, 1), padding=(3, 0), act='relu')
conv2 = self.conv_bn_layer(
conv2, 96, 3, act='relu', name="inception_stem2_3x3_2")
conv2, 64, (1, 7), padding=(0, 3), act='relu')
conv2 = self.conv_bn_layer(conv2, 96, 3, act='relu')
concat = fluid.layers.concat([conv1, conv2], axis=1)
conv1 = self.conv_bn_layer(
concat, 192, 3, stride=2, act='relu', name="inception_stem3_3x3_s2")
conv1 = self.conv_bn_layer(concat, 192, 3, stride=2, act='relu')
pool1 = fluid.layers.pool2d(
input=concat, pool_size=3, pool_stride=2, pool_type='max')
......@@ -132,207 +100,105 @@ class InceptionV4():
return concat
def inceptionA(self, data, name=None):
def inceptionA(self, data):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(
pool1, 96, 1, act='relu', name="inception_a" + name + "_1x1")
conv1 = self.conv_bn_layer(pool1, 96, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 96, 1, act='relu', name="inception_a" + name + "_1x1_2")
conv2 = self.conv_bn_layer(data, 96, 1, act='relu')
conv3 = self.conv_bn_layer(
data, 64, 1, act='relu', name="inception_a" + name + "_3x3_reduce")
conv3 = self.conv_bn_layer(
conv3,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3")
conv3 = self.conv_bn_layer(data, 64, 1, act='relu')
conv3 = self.conv_bn_layer(conv3, 96, 3, padding=1, act='relu')
conv4 = self.conv_bn_layer(
data,
64,
1,
act='relu',
name="inception_a" + name + "_3x3_2_reduce")
conv4 = self.conv_bn_layer(
conv4,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3_2")
conv4 = self.conv_bn_layer(
conv4,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3_3")
conv4 = self.conv_bn_layer(data, 64, 1, act='relu')
conv4 = self.conv_bn_layer(conv4, 96, 3, padding=1, act='relu')
conv4 = self.conv_bn_layer(conv4, 96, 3, padding=1, act='relu')
concat = fluid.layers.concat([conv1, conv2, conv3, conv4], axis=1)
return concat
def reductionA(self, data, name=None):
def reductionA(self, data):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(
data, 384, 3, stride=2, act='relu', name="reduction_a_3x3")
conv2 = self.conv_bn_layer(data, 384, 3, stride=2, act='relu')
conv3 = self.conv_bn_layer(
data, 192, 1, act='relu', name="reduction_a_3x3_2_reduce")
conv3 = self.conv_bn_layer(
conv3, 224, 3, padding=1, act='relu', name="reduction_a_3x3_2")
conv3 = self.conv_bn_layer(
conv3, 256, 3, stride=2, act='relu', name="reduction_a_3x3_3")
conv3 = self.conv_bn_layer(data, 192, 1, act='relu')
conv3 = self.conv_bn_layer(conv3, 224, 3, padding=1, act='relu')
conv3 = self.conv_bn_layer(conv3, 256, 3, stride=2, act='relu')
concat = fluid.layers.concat([pool1, conv2, conv3], axis=1)
return concat
def inceptionB(self, data, name=None):
def inceptionB(self, data):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(
pool1, 128, 1, act='relu', name="inception_b" + name + "_1x1")
conv1 = self.conv_bn_layer(pool1, 128, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 384, 1, act='relu', name="inception_b" + name + "_1x1_2")
conv2 = self.conv_bn_layer(data, 384, 1, act='relu')
conv3 = self.conv_bn_layer(data, 192, 1, act='relu')
conv3 = self.conv_bn_layer(
data, 192, 1, act='relu', name="inception_b" + name + "_1x7_reduce")
conv3 = self.conv_bn_layer(
conv3,
224, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7")
conv3, 224, (1, 7), padding=(0, 3), act='relu')
conv3 = self.conv_bn_layer(
conv3,
256, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1")
conv3, 256, (7, 1), padding=(3, 0), act='relu')
conv4 = self.conv_bn_layer(data, 192, 1, act='relu')
conv4 = self.conv_bn_layer(
data,
192,
1,
act='relu',
name="inception_b" + name + "_7x1_2_reduce")
conv4 = self.conv_bn_layer(
conv4,
192, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7_2")
conv4, 192, (1, 7), padding=(0, 3), act='relu')
conv4 = self.conv_bn_layer(
conv4,
224, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1_2")
conv4, 224, (7, 1), padding=(3, 0), act='relu')
conv4 = self.conv_bn_layer(
conv4,
224, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7_3")
conv4, 224, (1, 7), padding=(0, 3), act='relu')
conv4 = self.conv_bn_layer(
conv4,
256, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1_3")
conv4, 256, (7, 1), padding=(3, 0), act='relu')
concat = fluid.layers.concat([conv1, conv2, conv3, conv4], axis=1)
return concat
def reductionB(self, data, name=None):
def reductionB(self, data):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(
data, 192, 1, act='relu', name="reduction_b_3x3_reduce")
conv2 = self.conv_bn_layer(
conv2, 192, 3, stride=2, act='relu', name="reduction_b_3x3")
conv2 = self.conv_bn_layer(data, 192, 1, act='relu')
conv2 = self.conv_bn_layer(conv2, 192, 3, stride=2, act='relu')
conv3 = self.conv_bn_layer(data, 256, 1, act='relu')
conv3 = self.conv_bn_layer(
data, 256, 1, act='relu', name="reduction_b_1x7_reduce")
conv3 = self.conv_bn_layer(
conv3,
256, (1, 7),
padding=(0, 3),
act='relu',
name="reduction_b_1x7")
conv3 = self.conv_bn_layer(
conv3,
320, (7, 1),
padding=(3, 0),
act='relu',
name="reduction_b_7x1")
conv3, 256, (1, 7), padding=(0, 3), act='relu')
conv3 = self.conv_bn_layer(
conv3, 320, 3, stride=2, act='relu', name="reduction_b_3x3_2")
conv3, 320, (7, 1), padding=(3, 0), act='relu')
conv3 = self.conv_bn_layer(conv3, 320, 3, stride=2, act='relu')
concat = fluid.layers.concat([pool1, conv2, conv3], axis=1)
return concat
def inceptionC(self, data, name=None):
def inceptionC(self, data):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(
pool1, 256, 1, act='relu', name="inception_c" + name + "_1x1")
conv1 = self.conv_bn_layer(pool1, 256, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 256, 1, act='relu', name="inception_c" + name + "_1x1_2")
conv2 = self.conv_bn_layer(data, 256, 1, act='relu')
conv3 = self.conv_bn_layer(
data, 384, 1, act='relu', name="inception_c" + name + "_1x1_3")
conv3 = self.conv_bn_layer(data, 384, 1, act='relu')
conv3_1 = self.conv_bn_layer(
conv3,
256, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3")
conv3, 256, (1, 3), padding=(0, 1), act='relu')
conv3_2 = self.conv_bn_layer(
conv3,
256, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1")
conv3, 256, (3, 1), padding=(1, 0), act='relu')
conv4 = self.conv_bn_layer(data, 384, 1, act='relu')
conv4 = self.conv_bn_layer(
data, 384, 1, act='relu', name="inception_c" + name + "_1x1_4")
conv4 = self.conv_bn_layer(
conv4,
448, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3_2")
conv4, 448, (1, 3), padding=(0, 1), act='relu')
conv4 = self.conv_bn_layer(
conv4,
512, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1_2")
conv4, 512, (3, 1), padding=(1, 0), act='relu')
conv4_1 = self.conv_bn_layer(
conv4,
256, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3_3")
conv4, 256, (1, 3), padding=(0, 1), act='relu')
conv4_2 = self.conv_bn_layer(
conv4,
256, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1_3")
conv4, 256, (3, 1), padding=(1, 0), act='relu')
concat = fluid.layers.concat(
[conv1, conv2, conv3_1, conv3_2, conv4_1, conv4_2], axis=1)
......
fluid/PaddleCV/image_classification/models_name/mobilenet.py → fluid/PaddleCV/image_classification/legacy/models/mobilenet.py
浏览文件 @ dc1b032d
......@@ -32,8 +32,7 @@ class MobileNet():
channels=3,
num_filters=int(32 * scale),
stride=2,
padding=1,
name="conv1")
padding=1)
# 56x56
input = self.depthwise_separable(
......@@ -42,8 +41,7 @@ class MobileNet():
num_filters2=64,
num_groups=32,
stride=1,
scale=scale,
name="conv2_1")
scale=scale)
input = self.depthwise_separable(
input,
......@@ -51,8 +49,7 @@ class MobileNet():
num_filters2=128,
num_groups=64,
stride=2,
scale=scale,
name="conv2_2")
scale=scale)
# 28x28
input = self.depthwise_separable(
......@@ -61,8 +58,7 @@ class MobileNet():
num_filters2=128,
num_groups=128,
stride=1,
scale=scale,
name="conv3_1")
scale=scale)
input = self.depthwise_separable(
input,
......@@ -70,8 +66,7 @@ class MobileNet():
num_filters2=256,
num_groups=128,
stride=2,
scale=scale,
name="conv3_2")
scale=scale)
# 14x14
input = self.depthwise_separable(
......@@ -80,8 +75,7 @@ class MobileNet():
num_filters2=256,
num_groups=256,
stride=1,
scale=scale,
name="conv4_1")
scale=scale)
input = self.depthwise_separable(
input,
......@@ -89,8 +83,7 @@ class MobileNet():
num_filters2=512,
num_groups=256,
stride=2,
scale=scale,
name="conv4_2")
scale=scale)
# 14x14
for i in range(5):
......@@ -100,8 +93,7 @@ class MobileNet():
num_filters2=512,
num_groups=512,
stride=1,
scale=scale,
name="conv5" + "_" + str(i + 1))
scale=scale)
# 7x7
input = self.depthwise_separable(
input,
......@@ -109,8 +101,7 @@ class MobileNet():
num_filters2=1024,
num_groups=512,
stride=2,
scale=scale,
name="conv5_6")
scale=scale)
input = self.depthwise_separable(
input,
......@@ -118,8 +109,7 @@ class MobileNet():
num_filters2=1024,
num_groups=1024,
stride=1,
scale=scale,
name="conv6")
scale=scale)
input = fluid.layers.pool2d(
input=input,
......@@ -130,9 +120,7 @@ class MobileNet():
output = fluid.layers.fc(input=input,
size=class_dim,
param_attr=ParamAttr(
initializer=MSRA(), name="fc7_weights"),
bias_attr=ParamAttr(name="fc7_offset"))
param_attr=ParamAttr(initializer=MSRA()))
return output
def conv_bn_layer(self,
......@@ -144,8 +132,7 @@ class MobileNet():
channels=None,
num_groups=1,
act='relu',
use_cudnn=True,
name=None):
use_cudnn=True):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -155,26 +142,12 @@ class MobileNet():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(
initializer=MSRA(), name=name + "_weights"),
param_attr=ParamAttr(initializer=MSRA()),
bias_attr=False)
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def depthwise_separable(self,
input,
num_filters1,
num_filters2,
num_groups,
stride,
scale,
name=None):
return fluid.layers.batch_norm(input=conv, act=act)
def depthwise_separable(self, input, num_filters1, num_filters2, num_groups,
stride, scale):
depthwise_conv = self.conv_bn_layer(
input=input,
filter_size=3,
......@@ -182,14 +155,12 @@ class MobileNet():
stride=stride,
padding=1,
num_groups=int(num_groups * scale),
use_cudnn=False,
name=name + "_dw")
use_cudnn=False)
pointwise_conv = self.conv_bn_layer(
input=depthwise_conv,
filter_size=1,
num_filters=int(num_filters2 * scale),
stride=1,
padding=0,
name=name + "_sep")
padding=0)
return pointwise_conv
fluid/PaddleCV/image_classification/models_name/mobilenet_v2.py → fluid/PaddleCV/image_classification/legacy/models/mobilenet_v2.py
浏览文件 @ dc1b032d
......@@ -36,40 +36,33 @@ class MobileNetV2():
(6, 320, 1, 1),
]
#conv1
input = self.conv_bn_layer(
input,
num_filters=int(32 * scale),
filter_size=3,
stride=2,
padding=1,
if_act=True,
name='conv1_1')
if_act=True)
# bottleneck sequences
i = 1
in_c = int(32 * scale)
for layer_setting in bottleneck_params_list:
t, c, n, s = layer_setting
i += 1
input = self.invresi_blocks(
input=input,
in_c=in_c,
t=t,
c=int(c * scale),
n=n,
s=s,
name='conv' + str(i))
s=s, )
in_c = int(c * scale)
#last_conv
input = self.conv_bn_layer(
input=input,
num_filters=int(1280 * scale) if scale > 1.0 else 1280,
filter_size=1,
stride=1,
padding=0,
if_act=True,
name='conv9')
if_act=True)
input = fluid.layers.pool2d(
input=input,
......@@ -80,8 +73,7 @@ class MobileNetV2():
output = fluid.layers.fc(input=input,
size=class_dim,
param_attr=ParamAttr(name='fc10_weights'),
bias_attr=ParamAttr(name='fc10_offset'))
param_attr=ParamAttr(initializer=MSRA()))
return output
def conv_bn_layer(self,
......@@ -92,9 +84,8 @@ class MobileNetV2():
padding,
channels=None,
num_groups=1,
if_act=True,
name=None,
use_cudnn=True):
use_cudnn=True,
if_act=True):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -104,15 +95,9 @@ class MobileNetV2():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(name=name + '_weights'),
param_attr=ParamAttr(initializer=MSRA()),
bias_attr=False)
bn_name = name + '_bn'
bn = fluid.layers.batch_norm(
input=conv,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
bn = fluid.layers.batch_norm(input=conv)
if if_act:
return fluid.layers.relu6(bn)
else:
......@@ -121,18 +106,10 @@ class MobileNetV2():
def shortcut(self, input, data_residual):
return fluid.layers.elementwise_add(input, data_residual)
def inverted_residual_unit(self,
input,
num_in_filter,
num_filters,
ifshortcut,
stride,
filter_size,
padding,
expansion_factor,
name=None):
def inverted_residual_unit(self, input, num_in_filter, num_filters,
ifshortcut, stride, filter_size, padding,
expansion_factor):
num_expfilter = int(round(num_in_filter * expansion_factor))
channel_expand = self.conv_bn_layer(
input=input,
num_filters=num_expfilter,
......@@ -140,9 +117,7 @@ class MobileNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True,
name=name + '_expand')
if_act=True)
bottleneck_conv = self.conv_bn_layer(
input=channel_expand,
num_filters=num_expfilter,
......@@ -151,9 +126,7 @@ class MobileNetV2():
padding=padding,
num_groups=num_expfilter,
if_act=True,
name=name + '_dwise',
use_cudnn=False)
linear_out = self.conv_bn_layer(
input=bottleneck_conv,
num_filters=num_filters,
......@@ -161,15 +134,14 @@ class MobileNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=False,
name=name + '_linear')
if_act=False)
if ifshortcut:
out = self.shortcut(input=input, data_residual=linear_out)
return out
else:
return linear_out
def invresi_blocks(self, input, in_c, t, c, n, s, name=None):
def invresi_blocks(self, input, in_c, t, c, n, s):
first_block = self.inverted_residual_unit(
input=input,
num_in_filter=in_c,
......@@ -178,8 +150,7 @@ class MobileNetV2():
stride=s,
filter_size=3,
padding=1,
expansion_factor=t,
name=name + '_1')
expansion_factor=t)
last_residual_block = first_block
last_c = c
......@@ -193,6 +164,5 @@ class MobileNetV2():
stride=1,
filter_size=3,
padding=1,
expansion_factor=t,
name=name + '_' + str(i + 1))
expansion_factor=t)
return last_residual_block
fluid/PaddleCV/image_classification/models_name/resnet.py → fluid/PaddleCV/image_classification/legacy/models/resnet.py
浏览文件 @ dc1b032d
......@@ -4,7 +4,6 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ["ResNet", "ResNet50", "ResNet101", "ResNet152"]
......@@ -41,12 +40,7 @@ class ResNet():
num_filters = [64, 128, 256, 512]
conv = self.conv_bn_layer(
input=input,
num_filters=64,
filter_size=7,
stride=2,
act='relu',
name="conv1")
input=input, num_filters=64, filter_size=7, stride=2, act='relu')
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
......@@ -56,18 +50,10 @@ class ResNet():
for block in range(len(depth)):
for i in range(depth[block]):
if layers in [101, 152] and block == 2:
if i == 0:
conv_name = "res" + str(block + 2) + "a"
else:
conv_name = "res" + str(block + 2) + "b" + str(i)
else:
conv_name = "res" + str(block + 2) + chr(97 + i)
conv = self.bottleneck_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
name=conv_name)
stride=2 if i == 0 and block != 0 else 1)
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
......@@ -85,8 +71,7 @@ class ResNet():
filter_size,
stride=1,
groups=1,
act=None,
name=None):
act=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -95,55 +80,31 @@ class ResNet():
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False,
name=name + '.conv2d.output.1')
if name == "conv1":
bn_name = "bn_" + name
else:
bn_name = "bn" + name[3:]
return fluid.layers.batch_norm(
input=conv,
act=act,
name=bn_name + '.output.1',
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance', )
def shortcut(self, input, ch_out, stride, name):
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
def shortcut(self, input, ch_out, stride):
ch_in = input.shape[1]
if ch_in != ch_out or stride != 1:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
return self.conv_bn_layer(input, ch_out, 1, stride)
else:
return input
def bottleneck_block(self, input, num_filters, stride, name):
def bottleneck_block(self, input, num_filters, stride):
conv0 = self.conv_bn_layer(
input=input,
num_filters=num_filters,
filter_size=1,
act='relu',
name=name + "_branch2a")
input=input, num_filters=num_filters, filter_size=1, act='relu')
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
stride=stride,
act='relu',
name=name + "_branch2b")
act='relu')
conv2 = self.conv_bn_layer(
input=conv1,
num_filters=num_filters * 4,
filter_size=1,
act=None,
name=name + "_branch2c")
input=conv1, num_filters=num_filters * 4, filter_size=1, act=None)
short = self.shortcut(
input, num_filters * 4, stride, name=name + "_branch1")
short = self.shortcut(input, num_filters * 4, stride)
return fluid.layers.elementwise_add(
x=short, y=conv2, act='relu', name=name + ".add.output.5")
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')
def ResNet50():
......
fluid/PaddleCV/image_classification/models_name/se_resnext.py → fluid/PaddleCV/image_classification/legacy/models/se_resnext.py
浏览文件 @ dc1b032d
......@@ -4,7 +4,6 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = [
"SE_ResNeXt", "SE_ResNeXt50_32x4d", "SE_ResNeXt101_32x4d",
......@@ -19,7 +18,7 @@ train_parameters = {
"learning_strategy": {
"name": "piecewise_decay",
"batch_size": 256,
"epochs": [40, 80, 100],
"epochs": [30, 60, 90],
"steps": [0.1, 0.01, 0.001, 0.0001]
}
}
......@@ -46,8 +45,7 @@ class SE_ResNeXt():
num_filters=64,
filter_size=7,
stride=2,
act='relu',
name='conv1', )
act='relu')
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
......@@ -65,8 +63,7 @@ class SE_ResNeXt():
num_filters=64,
filter_size=7,
stride=2,
act='relu',
name="conv1", )
act='relu')
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
......@@ -84,94 +81,67 @@ class SE_ResNeXt():
num_filters=64,
filter_size=3,
stride=2,
act='relu',
name='conv1')
act='relu')
conv = self.conv_bn_layer(
input=conv,
num_filters=64,
filter_size=3,
stride=1,
act='relu',
name='conv2')
input=conv, num_filters=64, filter_size=3, stride=1, act='relu')
conv = self.conv_bn_layer(
input=conv,
num_filters=128,
filter_size=3,
stride=1,
act='relu',
name='conv3')
act='relu')
conv = fluid.layers.pool2d(
input=conv, pool_size=3, pool_stride=2, pool_padding=1, \
pool_type='max')
n = 1 if layers == 50 or layers == 101 else 3
for block in range(len(depth)):
n += 1
for i in range(depth[block]):
conv = self.bottleneck_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
cardinality=cardinality,
reduction_ratio=reduction_ratio,
name=str(n) + '_' + str(i + 1))
reduction_ratio=reduction_ratio)
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
drop = fluid.layers.dropout(
x=pool, dropout_prob=0.5, seed=self.params['dropout_seed'])
stdv = 1.0 / math.sqrt(drop.shape[1] * 1.0)
out = fluid.layers.fc(
input=drop,
out = fluid.layers.fc(input=drop,
size=class_dim,
param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv,
stdv)))
return out
def shortcut(self, input, ch_out, stride, name):
def shortcut(self, input, ch_out, stride):
ch_in = input.shape[1]
if ch_in != ch_out or stride != 1:
filter_size = 1
return self.conv_bn_layer(
input, ch_out, filter_size, stride, name='conv' + name + '_prj')
return self.conv_bn_layer(input, ch_out, filter_size, stride)
else:
return input
def bottleneck_block(self,
input,
num_filters,
stride,
cardinality,
reduction_ratio,
name=None):
def bottleneck_block(self, input, num_filters, stride, cardinality,
reduction_ratio):
conv0 = self.conv_bn_layer(
input=input,
num_filters=num_filters,
filter_size=1,
act='relu',
name='conv' + name + '_x1')
input=input, num_filters=num_filters, filter_size=1, act='relu')
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
stride=stride,
groups=cardinality,
act='relu',
name='conv' + name + '_x2')
act='relu')
conv2 = self.conv_bn_layer(
input=conv1,
num_filters=num_filters * 2,
filter_size=1,
act=None,
name='conv' + name + '_x3')
input=conv1, num_filters=num_filters * 2, filter_size=1, act=None)
scale = self.squeeze_excitation(
input=conv2,
num_channels=num_filters * 2,
reduction_ratio=reduction_ratio,
name='fc' + name)
reduction_ratio=reduction_ratio)
short = self.shortcut(input, num_filters * 2, stride, name=name)
short = self.shortcut(input, num_filters * 2, stride)
return fluid.layers.elementwise_add(x=short, y=scale, act='relu')
......@@ -181,8 +151,7 @@ class SE_ResNeXt():
filter_size,
stride=1,
groups=1,
act=None,
name=None):
act=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -191,42 +160,26 @@ class SE_ResNeXt():
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
bias_attr=False,
param_attr=ParamAttr(name=name + '_weights'), )
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
def squeeze_excitation(self,
input,
num_channels,
reduction_ratio,
name=None):
def squeeze_excitation(self, input, num_channels, reduction_ratio):
pool = fluid.layers.pool2d(
input=input, pool_size=0, pool_type='avg', global_pooling=True)
stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
squeeze = fluid.layers.fc(
input=pool,
squeeze = fluid.layers.fc(input=pool,
size=num_channels // reduction_ratio,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + '_sqz_weights'),
bias_attr=ParamAttr(name=name + '_sqz_offset'))
initializer=fluid.initializer.Uniform(
-stdv, stdv)))
stdv = 1.0 / math.sqrt(squeeze.shape[1] * 1.0)
excitation = fluid.layers.fc(
input=squeeze,
excitation = fluid.layers.fc(input=squeeze,
size=num_channels,
act='sigmoid',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + '_exc_weights'),
bias_attr=ParamAttr(name=name + '_exc_offset'))
initializer=fluid.initializer.Uniform(
-stdv, stdv)))
scale = fluid.layers.elementwise_mul(x=input, y=excitation, axis=0)
return scale
......
fluid/PaddleCV/image_classification/models_name/shufflenet_v2.py → fluid/PaddleCV/image_classification/legacy/models/shufflenet_v2.py
浏览文件 @ dc1b032d
......@@ -52,8 +52,7 @@ class ShuffleNetV2():
filter_size=3,
num_filters=input_channel,
padding=1,
stride=2,
name='stage1_conv')
stride=2)
pool1 = fluid.layers.pool2d(
input=conv1,
pool_size=3,
......@@ -71,35 +70,30 @@ class ShuffleNetV2():
input=conv,
num_filters=output_channel,
stride=2,
benchmodel=2,
name=str(idxstage + 2) + '_' + str(i + 1))
benchmodel=2)
else:
conv = self.inverted_residual_unit(
input=conv,
num_filters=output_channel,
stride=1,
benchmodel=1,
name=str(idxstage + 2) + '_' + str(i + 1))
benchmodel=1)
conv_last = self.conv_bn_layer(
input=conv,
filter_size=1,
num_filters=stage_out_channels[-1],
padding=0,
stride=1,
name='conv5')
stride=1)
pool_last = fluid.layers.pool2d(
input=conv_last,
pool_size=7,
pool_stride=1,
pool_stride=7,
pool_padding=0,
pool_type='avg')
output = fluid.layers.fc(input=pool_last,
size=class_dim,
param_attr=ParamAttr(
initializer=MSRA(), name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
param_attr=ParamAttr(initializer=MSRA()))
return output
def conv_bn_layer(self,
......@@ -110,9 +104,7 @@ class ShuffleNetV2():
padding,
num_groups=1,
use_cudnn=True,
if_act=True,
name=None):
# print(num_groups)
if_act=True):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -122,25 +114,12 @@ class ShuffleNetV2():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(
initializer=MSRA(), name=name + '_weights'),
param_attr=ParamAttr(initializer=MSRA()),
bias_attr=False)
bn_name = name + '_bn'
if if_act:
return fluid.layers.batch_norm(
input=conv,
act='relu',
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
return fluid.layers.batch_norm(input=conv, act='relu')
else:
return fluid.layers.batch_norm(
input=conv,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
return fluid.layers.batch_norm(input=conv)
def channel_shuffle(self, x, groups):
batchsize, num_channels, height, width = x.shape[0], x.shape[
......@@ -159,12 +138,7 @@ class ShuffleNetV2():
return x
def inverted_residual_unit(self,
input,
num_filters,
stride,
benchmodel,
name=None):
def inverted_residual_unit(self, input, num_filters, stride, benchmodel):
assert stride in [1, 2], \
"supported stride are {} but your stride is {}".format([1,2], stride)
......@@ -176,8 +150,6 @@ class ShuffleNetV2():
input,
num_or_sections=[input.shape[1] // 2, input.shape[1] // 2],
dim=1)
# x1 = input[:, :(input.shape[1]//2), :, :]
# x2 = input[:, (input.shape[1]//2):, :, :]
conv_pw = self.conv_bn_layer(
input=x2,
......@@ -186,8 +158,7 @@ class ShuffleNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True,
name='stage_' + name + '_conv1')
if_act=True)
conv_dw = self.conv_bn_layer(
input=conv_pw,
......@@ -196,8 +167,7 @@ class ShuffleNetV2():
stride=stride,
padding=1,
num_groups=oup_inc,
if_act=False,
name='stage_' + name + '_conv2')
if_act=False)
conv_linear = self.conv_bn_layer(
input=conv_dw,
......@@ -206,63 +176,57 @@ class ShuffleNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True,
name='stage_' + name + '_conv3')
if_act=True)
out = fluid.layers.concat([x1, conv_linear], axis=1)
else:
#branch1
conv_dw_1 = self.conv_bn_layer(
conv_dw = self.conv_bn_layer(
input=input,
num_filters=inp,
filter_size=3,
stride=stride,
padding=1,
num_groups=inp,
if_act=False,
name='stage_' + name + '_conv4')
if_act=False)
conv_linear_1 = self.conv_bn_layer(
input=conv_dw_1,
input=conv_dw,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True,
name='stage_' + name + '_conv5')
if_act=True)
#branch2
conv_pw_2 = self.conv_bn_layer(
conv_pw = self.conv_bn_layer(
input=input,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True,
name='stage_' + name + '_conv1')
if_act=True)
conv_dw_2 = self.conv_bn_layer(
input=conv_pw_2,
conv_dw = self.conv_bn_layer(
input=conv_pw,
num_filters=oup_inc,
filter_size=3,
stride=stride,
padding=1,
num_groups=oup_inc,
if_act=False,
name='stage_' + name + '_conv2')
if_act=False)
conv_linear_2 = self.conv_bn_layer(
input=conv_dw_2,
input=conv_dw,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True,
name='stage_' + name + '_conv3')
if_act=True)
out = fluid.layers.concat([conv_linear_1, conv_linear_2], axis=1)
return self.channel_shuffle(out, 2)
......
fluid/PaddleCV/image_classification/models_name/vgg.py → fluid/PaddleCV/image_classification/legacy/models/vgg.py
浏览文件 @ dc1b032d
......@@ -36,37 +36,42 @@ class VGGNet():
"supported layers are {} but input layer is {}".format(vgg_spec.keys(), layers)
nums = vgg_spec[layers]
conv1 = self.conv_block(input, 64, nums[0], name="conv1_")
conv2 = self.conv_block(conv1, 128, nums[1], name="conv2_")
conv3 = self.conv_block(conv2, 256, nums[2], name="conv3_")
conv4 = self.conv_block(conv3, 512, nums[3], name="conv4_")
conv5 = self.conv_block(conv4, 512, nums[4], name="conv5_")
conv1 = self.conv_block(input, 64, nums[0])
conv2 = self.conv_block(conv1, 128, nums[1])
conv3 = self.conv_block(conv2, 256, nums[2])
conv4 = self.conv_block(conv3, 512, nums[3])
conv5 = self.conv_block(conv4, 512, nums[4])
fc_dim = 4096
fc_name = ["fc6", "fc7", "fc8"]
fc1 = fluid.layers.fc(
input=conv5,
size=fc_dim,
act='relu',
param_attr=fluid.param_attr.ParamAttr(name=fc_name[0] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[0] + "_offset"))
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
fc1 = fluid.layers.dropout(x=fc1, dropout_prob=0.5)
fc2 = fluid.layers.fc(
input=fc1,
size=fc_dim,
act='relu',
param_attr=fluid.param_attr.ParamAttr(name=fc_name[1] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[1] + "_offset"))
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
fc2 = fluid.layers.dropout(x=fc2, dropout_prob=0.5)
out = fluid.layers.fc(
input=fc2,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(name=fc_name[2] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[2] + "_offset"))
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
return out
def conv_block(self, input, num_filter, groups, name=None):
def conv_block(self, input, num_filter, groups):
conv = input
for i in range(groups):
conv = fluid.layers.conv2d(
......@@ -77,9 +82,9 @@ class VGGNet():
padding=1,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
name=name + str(i + 1) + "_weights"),
initializer=fluid.initializer.Normal(scale=0.01)),
bias_attr=fluid.param_attr.ParamAttr(
name=name + str(i + 1) + "_offset"))
initializer=fluid.initializer.Constant(value=0.0)))
return fluid.layers.pool2d(
input=conv, pool_size=2, pool_type='max', pool_stride=2)
......
fluid/PaddleCV/image_classification/models/__init__.py
浏览文件 @ dc1b032d
......@@ -3,10 +3,10 @@ from .mobilenet import MobileNet
from .mobilenet_v2 import MobileNetV2
from .googlenet import GoogleNet
from .vgg import VGG11, VGG13, VGG16, VGG19
from .resnet import ResNet50, ResNet101, ResNet152
from .resnet import ResNet18, ResNet34, ResNet50, ResNet101, ResNet152
from .resnet_dist import DistResNet
from .inception_v4 import InceptionV4
from .se_resnext import SE_ResNeXt50_32x4d, SE_ResNeXt101_32x4d, SE_ResNeXt152_32x4d
from .dpn import DPN68, DPN92, DPN98, DPN107, DPN131
from .shufflenet_v2 import ShuffleNetV2_x0_5, ShuffleNetV2_x1_0, ShuffleNetV2_x1_5, ShuffleNetV2_x2_0
from .shufflenet_v2 import ShuffleNetV2, ShuffleNetV2_x0_5_swish, ShuffleNetV2_x1_0_swish, ShuffleNetV2_x1_5_swish, ShuffleNetV2_x2_0_swish, ShuffleNetV2_x8_0_swish
from .fast_imagenet import FastImageNet
fluid/PaddleCV/image_classification/models/alexnet.py
浏览文件 @ dc1b032d
......@@ -26,6 +26,9 @@ class AlexNet():
def net(self, input, class_dim=1000):
stdv = 1.0 / math.sqrt(input.shape[1] * 11 * 11)
layer_name = [
"conv1", "conv2", "conv3", "conv4", "conv5", "fc6", "fc7", "fc8"
]
conv1 = fluid.layers.conv2d(
input=input,
num_filters=64,
......@@ -35,9 +38,11 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[0] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[0] + "_weights"))
pool1 = fluid.layers.pool2d(
input=conv1,
pool_size=3,
......@@ -55,9 +60,11 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[1] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[1] + "_weights"))
pool2 = fluid.layers.pool2d(
input=conv2,
pool_size=3,
......@@ -75,9 +82,11 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[2] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[2] + "_weights"))
stdv = 1.0 / math.sqrt(conv3.shape[1] * 3 * 3)
conv4 = fluid.layers.conv2d(
......@@ -89,9 +98,11 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[3] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[3] + "_weights"))
stdv = 1.0 / math.sqrt(conv4.shape[1] * 3 * 3)
conv5 = fluid.layers.conv2d(
......@@ -103,9 +114,11 @@ class AlexNet():
groups=1,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[4] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[4] + "_weights"))
pool5 = fluid.layers.pool2d(
input=conv5,
pool_size=3,
......@@ -114,36 +127,42 @@ class AlexNet():
pool_type='max')
drop6 = fluid.layers.dropout(x=pool5, dropout_prob=0.5)
stdv = 1.0 / math.sqrt(drop6.shape[1] * drop6.shape[2] *
drop6.shape[3] * 1.0)
fc6 = fluid.layers.fc(
input=drop6,
size=4096,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[5] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[5] + "_weights"))
drop7 = fluid.layers.dropout(x=fc6, dropout_prob=0.5)
stdv = 1.0 / math.sqrt(drop7.shape[1] * 1.0)
fc7 = fluid.layers.fc(
input=drop7,
size=4096,
act='relu',
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[6] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[6] + "_weights"))
stdv = 1.0 / math.sqrt(fc7.shape[1] * 1.0)
out = fluid.layers.fc(
input=fc7,
size=class_dim,
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)),
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[7] + "_offset"),
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=layer_name[7] + "_weights"))
return out
fluid/PaddleCV/image_classification/models/dpn.py
浏览文件 @ dc1b032d
......@@ -7,6 +7,7 @@ import time
import sys
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ["DPN", "DPN68", "DPN92", "DPN98", "DPN107", "DPN131"]
......@@ -52,16 +53,30 @@ class DPN(object):
padding=init_padding,
groups=1,
act=None,
bias_attr=False)
bias_attr=False,
name="conv1",
param_attr=ParamAttr(name="conv1_weights"), )
conv1_x_1 = fluid.layers.batch_norm(
input=conv1_x_1, act='relu', is_test=False)
input=conv1_x_1,
act='relu',
is_test=False,
name="conv1_bn",
param_attr=ParamAttr(name='conv1_bn_scale'),
bias_attr=ParamAttr('conv1_bn_offset'),
moving_mean_name='conv1_bn_mean',
moving_variance_name='conv1_bn_variance', )
convX_x_x = fluid.layers.pool2d(
input=conv1_x_1,
pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
pool_type='max',
name="pool1")
#conv2 - conv5
match_list, num = [], 0
for gc in range(4):
bw = bws[gc]
inc = inc_sec[gc]
......@@ -69,32 +84,46 @@ class DPN(object):
if gc == 0:
_type1 = 'proj'
_type2 = 'normal'
match = 1
else:
_type1 = 'down'
_type2 = 'normal'
convX_x_x = self.dual_path_factory(convX_x_x, R, R, bw, inc, G,
_type1)
match = match + k_sec[gc - 1]
match_list.append(match)
convX_x_x = self.dual_path_factory(
convX_x_x, R, R, bw, inc, G, _type1, name="dpn" + str(match))
for i_ly in range(2, k_sec[gc] + 1):
convX_x_x = self.dual_path_factory(convX_x_x, R, R, bw, inc, G,
_type2)
num += 1
if num in match_list:
num += 1
convX_x_x = self.dual_path_factory(
convX_x_x, R, R, bw, inc, G, _type2, name="dpn" + str(num))
conv5_x_x = fluid.layers.concat(convX_x_x, axis=1)
conv5_x_x = fluid.layers.batch_norm(
input=conv5_x_x, act='relu', is_test=False)
input=conv5_x_x,
act='relu',
is_test=False,
name="final_concat_bn",
param_attr=ParamAttr(name='final_concat_bn_scale'),
bias_attr=ParamAttr('final_concat_bn_offset'),
moving_mean_name='final_concat_bn_mean',
moving_variance_name='final_concat_bn_variance', )
pool5 = fluid.layers.pool2d(
input=conv5_x_x,
pool_size=7,
pool_stride=1,
pool_padding=0,
pool_type='avg')
pool_type='avg', )
#stdv = 1.0 / math.sqrt(pool5.shape[1] * 1.0)
stdv = 0.01
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
fc6 = fluid.layers.fc(input=pool5,
size=class_dim,
param_attr=param_attr)
param_attr=param_attr,
name="fc6")
return fc6
......@@ -172,7 +201,8 @@ class DPN(object):
num_1x1_c,
inc,
G,
_type='normal'):
_type='normal',
name=None):
kw = 3
kh = 3
pw = (kw - 1) // 2
......@@ -201,35 +231,50 @@ class DPN(object):
num_filter=(num_1x1_c + 2 * inc),
kernel=(1, 1),
pad=(0, 0),
stride=(key_stride, key_stride))
stride=(key_stride, key_stride),
name=name + "_match")
data_o1, data_o2 = fluid.layers.split(
c1x1_w, num_or_sections=[num_1x1_c, 2 * inc], dim=1)
c1x1_w,
num_or_sections=[num_1x1_c, 2 * inc],
dim=1,
name=name + "_match_conv_Slice")
else:
data_o1 = data[0]
data_o2 = data[1]
# MAIN
c1x1_a = self.bn_ac_conv(
data=data_in, num_filter=num_1x1_a, kernel=(1, 1), pad=(0, 0))
data=data_in,
num_filter=num_1x1_a,
kernel=(1, 1),
pad=(0, 0),
name=name + "_conv1")
c3x3_b = self.bn_ac_conv(
data=c1x1_a,
num_filter=num_3x3_b,
kernel=(kw, kh),
pad=(pw, ph),
stride=(key_stride, key_stride),
num_group=G)
num_group=G,
name=name + "_conv2")
c1x1_c = self.bn_ac_conv(
data=c3x3_b,
num_filter=(num_1x1_c + inc),
kernel=(1, 1),
pad=(0, 0))
pad=(0, 0),
name=name + "_conv3")
c1x1_c1, c1x1_c2 = fluid.layers.split(
c1x1_c, num_or_sections=[num_1x1_c, inc], dim=1)
c1x1_c,
num_or_sections=[num_1x1_c, inc],
dim=1,
name=name + "_conv3_Slice")
# OUTPUTS
summ = fluid.layers.elementwise_add(x=data_o1, y=c1x1_c1)
dense = fluid.layers.concat([data_o2, c1x1_c2], axis=1)
summ = fluid.layers.elementwise_add(
x=data_o1, y=c1x1_c1, name=name + "_elewise")
dense = fluid.layers.concat(
[data_o2, c1x1_c2], axis=1, name=name + "_concat")
return [summ, dense]
......@@ -239,8 +284,17 @@ class DPN(object):
kernel,
pad,
stride=(1, 1),
num_group=1):
bn_ac = fluid.layers.batch_norm(input=data, act='relu', is_test=False)
num_group=1,
name=None):
bn_ac = fluid.layers.batch_norm(
input=data,
act='relu',
is_test=False,
name=name + '.output.1',
param_attr=ParamAttr(name=name + '_bn_scale'),
bias_attr=ParamAttr(name + '_bn_offset'),
moving_mean_name=name + '_bn_mean',
moving_variance_name=name + '_bn_variance', )
bn_ac_conv = fluid.layers.conv2d(
input=bn_ac,
num_filters=num_filter,
......@@ -249,7 +303,8 @@ class DPN(object):
padding=pad,
groups=num_group,
act=None,
bias_attr=False)
bias_attr=False,
param_attr=ParamAttr(name=name + "_weights"))
return bn_ac_conv
......@@ -259,7 +314,7 @@ def DPN68():
def DPN92():
model = DPN(layers=92)
onvodel = DPN(layers=92)
return model
......
fluid/PaddleCV/image_classification/models/googlenet.py
浏览文件 @ dc1b032d
......@@ -3,6 +3,7 @@ from __future__ import division
from __future__ import print_function
import paddle
import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
__all__ = ['GoogleNet']
......@@ -29,11 +30,13 @@ class GoogleNet():
filter_size,
stride=1,
groups=1,
act=None):
act=None,
name=None):
channels = input.shape[1]
stdv = (3.0 / (filter_size**2 * channels))**0.5
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
param_attr = ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + "_weights")
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -43,43 +46,63 @@ class GoogleNet():
groups=groups,
act=act,
param_attr=param_attr,
bias_attr=False)
bias_attr=False,
name=name)
return conv
def xavier(self, channels, filter_size):
def xavier(self, channels, filter_size, name):
stdv = (3.0 / (filter_size**2 * channels))**0.5
param_attr = fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv))
param_attr = ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + "_weights")
return param_attr
def inception(self, name, input, channels, filter1, filter3R, filter3,
filter5R, filter5, proj):
def inception(self,
input,
channels,
filter1,
filter3R,
filter3,
filter5R,
filter5,
proj,
name=None):
conv1 = self.conv_layer(
input=input, num_filters=filter1, filter_size=1, stride=1, act=None)
input=input,
num_filters=filter1,
filter_size=1,
stride=1,
act=None,
name="inception_" + name + "_1x1")
conv3r = self.conv_layer(
input=input,
num_filters=filter3R,
filter_size=1,
stride=1,
act=None)
act=None,
name="inception_" + name + "_3x3_reduce")
conv3 = self.conv_layer(
input=conv3r,
num_filters=filter3,
filter_size=3,
stride=1,
act=None)
act=None,
name="inception_" + name + "_3x3")
conv5r = self.conv_layer(
input=input,
num_filters=filter5R,
filter_size=1,
stride=1,
act=None)
act=None,
name="inception_" + name + "_5x5_reduce")
conv5 = self.conv_layer(
input=conv5r,
num_filters=filter5,
filter_size=5,
stride=1,
act=None)
act=None,
name="inception_" + name + "_5x5")
pool = fluid.layers.pool2d(
input=input,
pool_size=3,
......@@ -87,81 +110,124 @@ class GoogleNet():
pool_padding=1,
pool_type='max')
convprj = fluid.layers.conv2d(
input=pool, filter_size=1, num_filters=proj, stride=1, padding=0)
input=pool,
filter_size=1,
num_filters=proj,
stride=1,
padding=0,
name="inception_" + name + "_3x3_proj",
param_attr=ParamAttr(
name="inception_" + name + "_3x3_proj_weights"),
bias_attr=False)
cat = fluid.layers.concat(input=[conv1, conv3, conv5, convprj], axis=1)
cat = fluid.layers.relu(cat)
return cat
def net(self, input, class_dim=1000):
conv = self.conv_layer(
input=input, num_filters=64, filter_size=7, stride=2, act=None)
input=input,
num_filters=64,
filter_size=7,
stride=2,
act=None,
name="conv1")
pool = fluid.layers.pool2d(
input=conv, pool_size=3, pool_type='max', pool_stride=2)
conv = self.conv_layer(
input=pool, num_filters=64, filter_size=1, stride=1, act=None)
input=pool,
num_filters=64,
filter_size=1,
stride=1,
act=None,
name="conv2_1x1")
conv = self.conv_layer(
input=conv, num_filters=192, filter_size=3, stride=1, act=None)
input=conv,
num_filters=192,
filter_size=3,
stride=1,
act=None,
name="conv2_3x3")
pool = fluid.layers.pool2d(
input=conv, pool_size=3, pool_type='max', pool_stride=2)
ince3a = self.inception("ince3a", pool, 192, 64, 96, 128, 16, 32, 32)
ince3b = self.inception("ince3b", ince3a, 256, 128, 128, 192, 32, 96,
64)
ince3a = self.inception(pool, 192, 64, 96, 128, 16, 32, 32, "ince3a")
ince3b = self.inception(ince3a, 256, 128, 128, 192, 32, 96, 64,
"ince3b")
pool3 = fluid.layers.pool2d(
input=ince3b, pool_size=3, pool_type='max', pool_stride=2)
ince4a = self.inception("ince4a", pool3, 480, 192, 96, 208, 16, 48, 64)
ince4b = self.inception("ince4b", ince4a, 512, 160, 112, 224, 24, 64,
64)
ince4c = self.inception("ince4c", ince4b, 512, 128, 128, 256, 24, 64,
64)
ince4d = self.inception("ince4d", ince4c, 512, 112, 144, 288, 32, 64,
64)
ince4e = self.inception("ince4e", ince4d, 528, 256, 160, 320, 32, 128,
128)
ince4a = self.inception(pool3, 480, 192, 96, 208, 16, 48, 64, "ince4a")
ince4b = self.inception(ince4a, 512, 160, 112, 224, 24, 64, 64,
"ince4b")
ince4c = self.inception(ince4b, 512, 128, 128, 256, 24, 64, 64,
"ince4c")
ince4d = self.inception(ince4c, 512, 112, 144, 288, 32, 64, 64,
"ince4d")
ince4e = self.inception(ince4d, 528, 256, 160, 320, 32, 128, 128,
"ince4e")
pool4 = fluid.layers.pool2d(
input=ince4e, pool_size=3, pool_type='max', pool_stride=2)
ince5a = self.inception("ince5a", pool4, 832, 256, 160, 320, 32, 128,
128)
ince5b = self.inception("ince5b", ince5a, 832, 384, 192, 384, 48, 128,
128)
ince5a = self.inception(pool4, 832, 256, 160, 320, 32, 128, 128,
"ince5a")
ince5b = self.inception(ince5a, 832, 384, 192, 384, 48, 128, 128,
"ince5b")
pool5 = fluid.layers.pool2d(
input=ince5b, pool_size=7, pool_type='avg', pool_stride=7)
dropout = fluid.layers.dropout(x=pool5, dropout_prob=0.4)
out = fluid.layers.fc(input=dropout,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1))
param_attr=self.xavier(1024, 1, "out"),
name="out",
bias_attr=ParamAttr(name="out_offset"))
pool_o1 = fluid.layers.pool2d(
input=ince4a, pool_size=5, pool_type='avg', pool_stride=3)
conv_o1 = self.conv_layer(
input=pool_o1, num_filters=128, filter_size=1, stride=1, act=None)
input=pool_o1,
num_filters=128,
filter_size=1,
stride=1,
act=None,
name="conv_o1")
fc_o1 = fluid.layers.fc(input=conv_o1,
size=1024,
act='relu',
param_attr=self.xavier(2048, 1))
param_attr=self.xavier(2048, 1, "fc_o1"),
name="fc_o1",
bias_attr=ParamAttr(name="fc_o1_offset"))
dropout_o1 = fluid.layers.dropout(x=fc_o1, dropout_prob=0.7)
out1 = fluid.layers.fc(input=dropout_o1,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1))
param_attr=self.xavier(1024, 1, "out1"),
name="out1",
bias_attr=ParamAttr(name="out1_offset"))
pool_o2 = fluid.layers.pool2d(
input=ince4d, pool_size=5, pool_type='avg', pool_stride=3)
conv_o2 = self.conv_layer(
input=pool_o2, num_filters=128, filter_size=1, stride=1, act=None)
input=pool_o2,
num_filters=128,
filter_size=1,
stride=1,
act=None,
name="conv_o2")
fc_o2 = fluid.layers.fc(input=conv_o2,
size=1024,
act='relu',
param_attr=self.xavier(2048, 1))
param_attr=self.xavier(2048, 1, "fc_o2"),
name="fc_o2",
bias_attr=ParamAttr(name="fc_o2_offset"))
dropout_o2 = fluid.layers.dropout(x=fc_o2, dropout_prob=0.7)
out2 = fluid.layers.fc(input=dropout_o2,
size=class_dim,
act='softmax',
param_attr=self.xavier(1024, 1))
param_attr=self.xavier(1024, 1, "out2"),
name="out2",
bias_attr=ParamAttr(name="out2_offset"))
# last fc layer is "out"
return out, out1, out2
fluid/PaddleCV/image_classification/models/inception_v4.py
浏览文件 @ dc1b032d
......@@ -4,6 +4,7 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ['InceptionV4']
......@@ -28,15 +29,15 @@ class InceptionV4():
x = self.inception_stem(input)
for i in range(4):
x = self.inceptionA(x)
x = self.inceptionA(x, name=str(i + 1))
x = self.reductionA(x)
for i in range(7):
x = self.inceptionB(x)
x = self.inceptionB(x, name=str(i + 1))
x = self.reductionB(x)
for i in range(3):
x = self.inceptionC(x)
x = self.inceptionC(x, name=str(i + 1))
pool = fluid.layers.pool2d(
input=x, pool_size=8, pool_type='avg', global_pooling=True)
......@@ -47,8 +48,12 @@ class InceptionV4():
out = fluid.layers.fc(
input=drop,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name="final_fc_weights"),
bias_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name="final_fc_offset"))
return out
def conv_bn_layer(self,
......@@ -58,7 +63,8 @@ class InceptionV4():
stride=1,
padding=0,
groups=1,
act='relu'):
act='relu',
name=None):
conv = fluid.layers.conv2d(
input=data,
num_filters=num_filters,
......@@ -67,32 +73,58 @@ class InceptionV4():
padding=padding,
groups=groups,
act=None,
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
def inception_stem(self, data):
conv = self.conv_bn_layer(data, 32, 3, stride=2, act='relu')
conv = self.conv_bn_layer(conv, 32, 3, act='relu')
conv = self.conv_bn_layer(conv, 64, 3, padding=1, act='relu')
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False,
name=name)
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
name=bn_name,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def inception_stem(self, data, name=None):
conv = self.conv_bn_layer(
data, 32, 3, stride=2, act='relu', name="conv1_3x3_s2")
conv = self.conv_bn_layer(conv, 32, 3, act='relu', name="conv2_3x3_s1")
conv = self.conv_bn_layer(
conv, 64, 3, padding=1, act='relu', name="conv3_3x3_s1")
pool1 = fluid.layers.pool2d(
input=conv, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(conv, 96, 3, stride=2, act='relu')
conv2 = self.conv_bn_layer(
conv, 96, 3, stride=2, act='relu', name="inception_stem1_3x3_s2")
concat = fluid.layers.concat([pool1, conv2], axis=1)
conv1 = self.conv_bn_layer(concat, 64, 1, act='relu')
conv1 = self.conv_bn_layer(conv1, 96, 3, act='relu')
conv1 = self.conv_bn_layer(
concat, 64, 1, act='relu', name="inception_stem2_3x3_reduce")
conv1 = self.conv_bn_layer(
conv1, 96, 3, act='relu', name="inception_stem2_3x3")
conv2 = self.conv_bn_layer(concat, 64, 1, act='relu')
conv2 = self.conv_bn_layer(
conv2, 64, (7, 1), padding=(3, 0), act='relu')
concat, 64, 1, act='relu', name="inception_stem2_1x7_reduce")
conv2 = self.conv_bn_layer(
conv2,
64, (7, 1),
padding=(3, 0),
act='relu',
name="inception_stem2_1x7")
conv2 = self.conv_bn_layer(
conv2,
64, (1, 7),
padding=(0, 3),
act='relu',
name="inception_stem2_7x1")
conv2 = self.conv_bn_layer(
conv2, 64, (1, 7), padding=(0, 3), act='relu')
conv2 = self.conv_bn_layer(conv2, 96, 3, act='relu')
conv2, 96, 3, act='relu', name="inception_stem2_3x3_2")
concat = fluid.layers.concat([conv1, conv2], axis=1)
conv1 = self.conv_bn_layer(concat, 192, 3, stride=2, act='relu')
conv1 = self.conv_bn_layer(
concat, 192, 3, stride=2, act='relu', name="inception_stem3_3x3_s2")
pool1 = fluid.layers.pool2d(
input=concat, pool_size=3, pool_stride=2, pool_type='max')
......@@ -100,105 +132,207 @@ class InceptionV4():
return concat
def inceptionA(self, data):
def inceptionA(self, data, name=None):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(pool1, 96, 1, act='relu')
conv1 = self.conv_bn_layer(
pool1, 96, 1, act='relu', name="inception_a" + name + "_1x1")
conv2 = self.conv_bn_layer(data, 96, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 96, 1, act='relu', name="inception_a" + name + "_1x1_2")
conv3 = self.conv_bn_layer(data, 64, 1, act='relu')
conv3 = self.conv_bn_layer(conv3, 96, 3, padding=1, act='relu')
conv3 = self.conv_bn_layer(
data, 64, 1, act='relu', name="inception_a" + name + "_3x3_reduce")
conv3 = self.conv_bn_layer(
conv3,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3")
conv4 = self.conv_bn_layer(data, 64, 1, act='relu')
conv4 = self.conv_bn_layer(conv4, 96, 3, padding=1, act='relu')
conv4 = self.conv_bn_layer(conv4, 96, 3, padding=1, act='relu')
conv4 = self.conv_bn_layer(
data,
64,
1,
act='relu',
name="inception_a" + name + "_3x3_2_reduce")
conv4 = self.conv_bn_layer(
conv4,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3_2")
conv4 = self.conv_bn_layer(
conv4,
96,
3,
padding=1,
act='relu',
name="inception_a" + name + "_3x3_3")
concat = fluid.layers.concat([conv1, conv2, conv3, conv4], axis=1)
return concat
def reductionA(self, data):
def reductionA(self, data, name=None):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(data, 384, 3, stride=2, act='relu')
conv2 = self.conv_bn_layer(
data, 384, 3, stride=2, act='relu', name="reduction_a_3x3")
conv3 = self.conv_bn_layer(data, 192, 1, act='relu')
conv3 = self.conv_bn_layer(conv3, 224, 3, padding=1, act='relu')
conv3 = self.conv_bn_layer(conv3, 256, 3, stride=2, act='relu')
conv3 = self.conv_bn_layer(
data, 192, 1, act='relu', name="reduction_a_3x3_2_reduce")
conv3 = self.conv_bn_layer(
conv3, 224, 3, padding=1, act='relu', name="reduction_a_3x3_2")
conv3 = self.conv_bn_layer(
conv3, 256, 3, stride=2, act='relu', name="reduction_a_3x3_3")
concat = fluid.layers.concat([pool1, conv2, conv3], axis=1)
return concat
def inceptionB(self, data):
def inceptionB(self, data, name=None):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(pool1, 128, 1, act='relu')
conv1 = self.conv_bn_layer(
pool1, 128, 1, act='relu', name="inception_b" + name + "_1x1")
conv2 = self.conv_bn_layer(data, 384, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 384, 1, act='relu', name="inception_b" + name + "_1x1_2")
conv3 = self.conv_bn_layer(data, 192, 1, act='relu')
conv3 = self.conv_bn_layer(
conv3, 224, (1, 7), padding=(0, 3), act='relu')
data, 192, 1, act='relu', name="inception_b" + name + "_1x7_reduce")
conv3 = self.conv_bn_layer(
conv3,
224, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7")
conv3 = self.conv_bn_layer(
conv3, 256, (7, 1), padding=(3, 0), act='relu')
conv3,
256, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1")
conv4 = self.conv_bn_layer(data, 192, 1, act='relu')
conv4 = self.conv_bn_layer(
conv4, 192, (1, 7), padding=(0, 3), act='relu')
data,
192,
1,
act='relu',
name="inception_b" + name + "_7x1_2_reduce")
conv4 = self.conv_bn_layer(
conv4,
192, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7_2")
conv4 = self.conv_bn_layer(
conv4, 224, (7, 1), padding=(3, 0), act='relu')
conv4,
224, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1_2")
conv4 = self.conv_bn_layer(
conv4, 224, (1, 7), padding=(0, 3), act='relu')
conv4,
224, (1, 7),
padding=(0, 3),
act='relu',
name="inception_b" + name + "_1x7_3")
conv4 = self.conv_bn_layer(
conv4, 256, (7, 1), padding=(3, 0), act='relu')
conv4,
256, (7, 1),
padding=(3, 0),
act='relu',
name="inception_b" + name + "_7x1_3")
concat = fluid.layers.concat([conv1, conv2, conv3, conv4], axis=1)
return concat
def reductionB(self, data):
def reductionB(self, data, name=None):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_stride=2, pool_type='max')
conv2 = self.conv_bn_layer(data, 192, 1, act='relu')
conv2 = self.conv_bn_layer(conv2, 192, 3, stride=2, act='relu')
conv2 = self.conv_bn_layer(
data, 192, 1, act='relu', name="reduction_b_3x3_reduce")
conv2 = self.conv_bn_layer(
conv2, 192, 3, stride=2, act='relu', name="reduction_b_3x3")
conv3 = self.conv_bn_layer(data, 256, 1, act='relu')
conv3 = self.conv_bn_layer(
conv3, 256, (1, 7), padding=(0, 3), act='relu')
data, 256, 1, act='relu', name="reduction_b_1x7_reduce")
conv3 = self.conv_bn_layer(
conv3,
256, (1, 7),
padding=(0, 3),
act='relu',
name="reduction_b_1x7")
conv3 = self.conv_bn_layer(
conv3,
320, (7, 1),
padding=(3, 0),
act='relu',
name="reduction_b_7x1")
conv3 = self.conv_bn_layer(
conv3, 320, (7, 1), padding=(3, 0), act='relu')
conv3 = self.conv_bn_layer(conv3, 320, 3, stride=2, act='relu')
conv3, 320, 3, stride=2, act='relu', name="reduction_b_3x3_2")
concat = fluid.layers.concat([pool1, conv2, conv3], axis=1)
return concat
def inceptionC(self, data):
def inceptionC(self, data, name=None):
pool1 = fluid.layers.pool2d(
input=data, pool_size=3, pool_padding=1, pool_type='avg')
conv1 = self.conv_bn_layer(pool1, 256, 1, act='relu')
conv1 = self.conv_bn_layer(
pool1, 256, 1, act='relu', name="inception_c" + name + "_1x1")
conv2 = self.conv_bn_layer(data, 256, 1, act='relu')
conv2 = self.conv_bn_layer(
data, 256, 1, act='relu', name="inception_c" + name + "_1x1_2")
conv3 = self.conv_bn_layer(data, 384, 1, act='relu')
conv3 = self.conv_bn_layer(
data, 384, 1, act='relu', name="inception_c" + name + "_1x1_3")
conv3_1 = self.conv_bn_layer(
conv3, 256, (1, 3), padding=(0, 1), act='relu')
conv3,
256, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3")
conv3_2 = self.conv_bn_layer(
conv3, 256, (3, 1), padding=(1, 0), act='relu')
conv3,
256, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1")
conv4 = self.conv_bn_layer(data, 384, 1, act='relu')
conv4 = self.conv_bn_layer(
conv4, 448, (1, 3), padding=(0, 1), act='relu')
data, 384, 1, act='relu', name="inception_c" + name + "_1x1_4")
conv4 = self.conv_bn_layer(
conv4,
448, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3_2")
conv4 = self.conv_bn_layer(
conv4, 512, (3, 1), padding=(1, 0), act='relu')
conv4,
512, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1_2")
conv4_1 = self.conv_bn_layer(
conv4, 256, (1, 3), padding=(0, 1), act='relu')
conv4,
256, (1, 3),
padding=(0, 1),
act='relu',
name="inception_c" + name + "_1x3_3")
conv4_2 = self.conv_bn_layer(
conv4, 256, (3, 1), padding=(1, 0), act='relu')
conv4,
256, (3, 1),
padding=(1, 0),
act='relu',
name="inception_c" + name + "_3x1_3")
concat = fluid.layers.concat(
[conv1, conv2, conv3_1, conv3_2, conv4_1, conv4_2], axis=1)
......
fluid/PaddleCV/image_classification/models/mobilenet.py
浏览文件 @ dc1b032d
......@@ -32,7 +32,8 @@ class MobileNet():
channels=3,
num_filters=int(32 * scale),
stride=2,
padding=1)
padding=1,
name="conv1")
# 56x56
input = self.depthwise_separable(
......@@ -41,7 +42,8 @@ class MobileNet():
num_filters2=64,
num_groups=32,
stride=1,
scale=scale)
scale=scale,
name="conv2_1")
input = self.depthwise_separable(
input,
......@@ -49,7 +51,8 @@ class MobileNet():
num_filters2=128,
num_groups=64,
stride=2,
scale=scale)
scale=scale,
name="conv2_2")
# 28x28
input = self.depthwise_separable(
......@@ -58,7 +61,8 @@ class MobileNet():
num_filters2=128,
num_groups=128,
stride=1,
scale=scale)
scale=scale,
name="conv3_1")
input = self.depthwise_separable(
input,
......@@ -66,7 +70,8 @@ class MobileNet():
num_filters2=256,
num_groups=128,
stride=2,
scale=scale)
scale=scale,
name="conv3_2")
# 14x14
input = self.depthwise_separable(
......@@ -75,7 +80,8 @@ class MobileNet():
num_filters2=256,
num_groups=256,
stride=1,
scale=scale)
scale=scale,
name="conv4_1")
input = self.depthwise_separable(
input,
......@@ -83,7 +89,8 @@ class MobileNet():
num_filters2=512,
num_groups=256,
stride=2,
scale=scale)
scale=scale,
name="conv4_2")
# 14x14
for i in range(5):
......@@ -93,7 +100,8 @@ class MobileNet():
num_filters2=512,
num_groups=512,
stride=1,
scale=scale)
scale=scale,
name="conv5" + "_" + str(i + 1))
# 7x7
input = self.depthwise_separable(
input,
......@@ -101,7 +109,8 @@ class MobileNet():
num_filters2=1024,
num_groups=512,
stride=2,
scale=scale)
scale=scale,
name="conv5_6")
input = self.depthwise_separable(
input,
......@@ -109,7 +118,8 @@ class MobileNet():
num_filters2=1024,
num_groups=1024,
stride=1,
scale=scale)
scale=scale,
name="conv6")
input = fluid.layers.pool2d(
input=input,
......@@ -120,7 +130,9 @@ class MobileNet():
output = fluid.layers.fc(input=input,
size=class_dim,
param_attr=ParamAttr(initializer=MSRA()))
param_attr=ParamAttr(
initializer=MSRA(), name="fc7_weights"),
bias_attr=ParamAttr(name="fc7_offset"))
return output
def conv_bn_layer(self,
......@@ -132,7 +144,8 @@ class MobileNet():
channels=None,
num_groups=1,
act='relu',
use_cudnn=True):
use_cudnn=True,
name=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -142,12 +155,26 @@ class MobileNet():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(initializer=MSRA()),
param_attr=ParamAttr(
initializer=MSRA(), name=name + "_weights"),
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
def depthwise_separable(self, input, num_filters1, num_filters2, num_groups,
stride, scale):
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def depthwise_separable(self,
input,
num_filters1,
num_filters2,
num_groups,
stride,
scale,
name=None):
depthwise_conv = self.conv_bn_layer(
input=input,
filter_size=3,
......@@ -155,12 +182,14 @@ class MobileNet():
stride=stride,
padding=1,
num_groups=int(num_groups * scale),
use_cudnn=False)
use_cudnn=False,
name=name + "_dw")
pointwise_conv = self.conv_bn_layer(
input=depthwise_conv,
filter_size=1,
num_filters=int(num_filters2 * scale),
stride=1,
padding=0)
padding=0,
name=name + "_sep")
return pointwise_conv
fluid/PaddleCV/image_classification/models/mobilenet_v2.py
浏览文件 @ dc1b032d
......@@ -36,33 +36,40 @@ class MobileNetV2():
(6, 320, 1, 1),
]
#conv1
input = self.conv_bn_layer(
input,
num_filters=int(32 * scale),
filter_size=3,
stride=2,
padding=1,
if_act=True)
if_act=True,
name='conv1_1')
# bottleneck sequences
i = 1
in_c = int(32 * scale)
for layer_setting in bottleneck_params_list:
t, c, n, s = layer_setting
i += 1
input = self.invresi_blocks(
input=input,
in_c=in_c,
t=t,
c=int(c * scale),
n=n,
s=s, )
s=s,
name='conv' + str(i))
in_c = int(c * scale)
#last_conv
input = self.conv_bn_layer(
input=input,
num_filters=int(1280 * scale) if scale > 1.0 else 1280,
filter_size=1,
stride=1,
padding=0,
if_act=True)
if_act=True,
name='conv9')
input = fluid.layers.pool2d(
input=input,
......@@ -73,7 +80,8 @@ class MobileNetV2():
output = fluid.layers.fc(input=input,
size=class_dim,
param_attr=ParamAttr(initializer=MSRA()))
param_attr=ParamAttr(name='fc10_weights'),
bias_attr=ParamAttr(name='fc10_offset'))
return output
def conv_bn_layer(self,
......@@ -84,8 +92,9 @@ class MobileNetV2():
padding,
channels=None,
num_groups=1,
use_cudnn=True,
if_act=True):
if_act=True,
name=None,
use_cudnn=True):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -95,9 +104,15 @@ class MobileNetV2():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(initializer=MSRA()),
param_attr=ParamAttr(name=name + '_weights'),
bias_attr=False)
bn = fluid.layers.batch_norm(input=conv)
bn_name = name + '_bn'
bn = fluid.layers.batch_norm(
input=conv,
param_attr=ParamAttr(name=bn_name + "_scale"),
bias_attr=ParamAttr(name=bn_name + "_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
if if_act:
return fluid.layers.relu6(bn)
else:
......@@ -106,10 +121,18 @@ class MobileNetV2():
def shortcut(self, input, data_residual):
return fluid.layers.elementwise_add(input, data_residual)
def inverted_residual_unit(self, input, num_in_filter, num_filters,
ifshortcut, stride, filter_size, padding,
expansion_factor):
def inverted_residual_unit(self,
input,
num_in_filter,
num_filters,
ifshortcut,
stride,
filter_size,
padding,
expansion_factor,
name=None):
num_expfilter = int(round(num_in_filter * expansion_factor))
channel_expand = self.conv_bn_layer(
input=input,
num_filters=num_expfilter,
......@@ -117,7 +140,9 @@ class MobileNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name=name + '_expand')
bottleneck_conv = self.conv_bn_layer(
input=channel_expand,
num_filters=num_expfilter,
......@@ -126,7 +151,9 @@ class MobileNetV2():
padding=padding,
num_groups=num_expfilter,
if_act=True,
name=name + '_dwise',
use_cudnn=False)
linear_out = self.conv_bn_layer(
input=bottleneck_conv,
num_filters=num_filters,
......@@ -134,14 +161,15 @@ class MobileNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=False)
if_act=False,
name=name + '_linear')
if ifshortcut:
out = self.shortcut(input=input, data_residual=linear_out)
return out
else:
return linear_out
def invresi_blocks(self, input, in_c, t, c, n, s):
def invresi_blocks(self, input, in_c, t, c, n, s, name=None):
first_block = self.inverted_residual_unit(
input=input,
num_in_filter=in_c,
......@@ -150,7 +178,8 @@ class MobileNetV2():
stride=s,
filter_size=3,
padding=1,
expansion_factor=t)
expansion_factor=t,
name=name + '_1')
last_residual_block = first_block
last_c = c
......@@ -164,5 +193,6 @@ class MobileNetV2():
stride=1,
filter_size=3,
padding=1,
expansion_factor=t)
expansion_factor=t,
name=name + '_' + str(i + 1))
return last_residual_block
fluid/PaddleCV/image_classification/models/resnet.py
浏览文件 @ dc1b032d
......@@ -4,8 +4,9 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = ["ResNet", "ResNet50", "ResNet101", "ResNet152"]
__all__ = ["ResNet", "ResNet18", "ResNet34", "ResNet50", "ResNet101", "ResNet152"]
train_parameters = {
"input_size": [3, 224, 224],
......@@ -27,11 +28,13 @@ class ResNet():
def net(self, input, class_dim=1000):
layers = self.layers
supported_layers = [50, 101, 152]
supported_layers = [18, 34, 50, 101, 152]
assert layers in supported_layers, \
"supported layers are {} but input layer is {}".format(supported_layers, layers)
if layers == 50:
if layers == 18:
depth = [2, 2, 2, 2]
elif layers == 34 or layers == 50:
depth = [3, 4, 6, 3]
elif layers == 101:
depth = [3, 4, 23, 3]
......@@ -40,20 +43,45 @@ class ResNet():
num_filters = [64, 128, 256, 512]
conv = self.conv_bn_layer(
input=input, num_filters=64, filter_size=7, stride=2, act='relu')
input=input, num_filters=64, filter_size=7, stride=2, act='relu',name="conv1")
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
if layers >= 50:
for block in range(len(depth)):
for i in range(depth[block]):
if layers in [101, 152] and block == 2:
if i == 0:
conv_name="res"+str(block+2)+"a"
else:
conv_name="res"+str(block+2)+"b"+str(i)
else:
conv_name="res"+str(block+2)+chr(97+i)
conv = self.bottleneck_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1)
stride=2 if i == 0 and block != 0 else 1, name=conv_name)
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
out = fluid.layers.fc(input=pool,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
else:
for block in range(len(depth)):
for i in range(depth[block]):
conv_name="res"+str(block+2)+chr(97+i)
conv = self.basic_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
is_first=block==i==0,
name=conv_name)
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
......@@ -61,8 +89,7 @@ class ResNet():
out = fluid.layers.fc(input=pool,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv,
stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv)))
return out
def conv_bn_layer(self,
......@@ -71,40 +98,73 @@ class ResNet():
filter_size,
stride=1,
groups=1,
act=None):
act=None,
name=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
padding=(filter_size - 1) / 2,
groups=groups,
act=None,
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
param_attr=ParamAttr(name=name + "_weights"),
bias_attr=False,
name=name + '.conv2d.output.1')
def shortcut(self, input, ch_out, stride):
if name == "conv1":
bn_name = "bn_" + name
else:
bn_name = "bn" + name[3:]
return fluid.layers.batch_norm(input=conv,
act=act,
name=bn_name+'.output.1',
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance',)
def shortcut(self, input, ch_out, stride, is_first, name):
ch_in = input.shape[1]
if ch_in != ch_out or stride != 1:
return self.conv_bn_layer(input, ch_out, 1, stride)
if ch_in != ch_out or stride != 1 or is_first == True:
return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
else:
return input
def bottleneck_block(self, input, num_filters, stride):
def bottleneck_block(self, input, num_filters, stride, name):
conv0 = self.conv_bn_layer(
input=input, num_filters=num_filters, filter_size=1, act='relu')
input=input, num_filters=num_filters, filter_size=1, act='relu',name=name+"_branch2a")
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
stride=stride,
act='relu')
act='relu',
name=name+"_branch2b")
conv2 = self.conv_bn_layer(
input=conv1, num_filters=num_filters * 4, filter_size=1, act=None)
input=conv1, num_filters=num_filters * 4, filter_size=1, act=None, name=name+"_branch2c")
short = self.shortcut(input, num_filters * 4, stride, is_first=False, name=name + "_branch1")
short = self.shortcut(input, num_filters * 4, stride)
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu',name=name+".add.output.5")
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')
def basic_block(self, input, num_filters, stride, is_first, name):
conv0 = self.conv_bn_layer(input=input, num_filters=num_filters, filter_size=3, act='relu', stride=stride,
name=name+"_branch2a")
conv1 = self.conv_bn_layer(input=conv0, num_filters=num_filters, filter_size=3, act=None,
name=name+"_branch2b")
short = self.shortcut(input, num_filters, stride, is_first, name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=conv1, act='relu')
def ResNet18():
model = ResNet(layers=18)
return model
def ResNet34():
model = ResNet(layers=34)
return model
def ResNet50():
......
fluid/PaddleCV/image_classification/models/se_resnext.py
浏览文件 @ dc1b032d
......@@ -4,6 +4,7 @@ from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr
__all__ = [
"SE_ResNeXt", "SE_ResNeXt50_32x4d", "SE_ResNeXt101_32x4d",
......@@ -18,7 +19,7 @@ train_parameters = {
"learning_strategy": {
"name": "piecewise_decay",
"batch_size": 256,
"epochs": [30, 60, 90],
"epochs": [40, 80, 100],
"steps": [0.1, 0.01, 0.001, 0.0001]
}
}
......@@ -45,7 +46,8 @@ class SE_ResNeXt():
num_filters=64,
filter_size=7,
stride=2,
act='relu')
act='relu',
name='conv1', )
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
......@@ -63,7 +65,8 @@ class SE_ResNeXt():
num_filters=64,
filter_size=7,
stride=2,
act='relu')
act='relu',
name="conv1", )
conv = fluid.layers.pool2d(
input=conv,
pool_size=3,
......@@ -81,67 +84,94 @@ class SE_ResNeXt():
num_filters=64,
filter_size=3,
stride=2,
act='relu')
act='relu',
name='conv1')
conv = self.conv_bn_layer(
input=conv, num_filters=64, filter_size=3, stride=1, act='relu')
input=conv,
num_filters=64,
filter_size=3,
stride=1,
act='relu',
name='conv2')
conv = self.conv_bn_layer(
input=conv,
num_filters=128,
filter_size=3,
stride=1,
act='relu')
act='relu',
name='conv3')
conv = fluid.layers.pool2d(
input=conv, pool_size=3, pool_stride=2, pool_padding=1, \
pool_type='max')
n = 1 if layers == 50 or layers == 101 else 3
for block in range(len(depth)):
n += 1
for i in range(depth[block]):
conv = self.bottleneck_block(
input=conv,
num_filters=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
cardinality=cardinality,
reduction_ratio=reduction_ratio)
reduction_ratio=reduction_ratio,
name=str(n) + '_' + str(i + 1))
pool = fluid.layers.pool2d(
input=conv, pool_size=7, pool_type='avg', global_pooling=True)
drop = fluid.layers.dropout(
x=pool, dropout_prob=0.5, seed=self.params['dropout_seed'])
stdv = 1.0 / math.sqrt(drop.shape[1] * 1.0)
out = fluid.layers.fc(input=drop,
out = fluid.layers.fc(
input=drop,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv,
stdv)))
param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv),
name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
return out
def shortcut(self, input, ch_out, stride):
def shortcut(self, input, ch_out, stride, name):
ch_in = input.shape[1]
if ch_in != ch_out or stride != 1:
filter_size = 1
return self.conv_bn_layer(input, ch_out, filter_size, stride)
return self.conv_bn_layer(
input, ch_out, filter_size, stride, name='conv' + name + '_prj')
else:
return input
def bottleneck_block(self, input, num_filters, stride, cardinality,
reduction_ratio):
def bottleneck_block(self,
input,
num_filters,
stride,
cardinality,
reduction_ratio,
name=None):
conv0 = self.conv_bn_layer(
input=input, num_filters=num_filters, filter_size=1, act='relu')
input=input,
num_filters=num_filters,
filter_size=1,
act='relu',
name='conv' + name + '_x1')
conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
stride=stride,
groups=cardinality,
act='relu')
act='relu',
name='conv' + name + '_x2')
conv2 = self.conv_bn_layer(
input=conv1, num_filters=num_filters * 2, filter_size=1, act=None)
input=conv1,
num_filters=num_filters * 2,
filter_size=1,
act=None,
name='conv' + name + '_x3')
scale = self.squeeze_excitation(
input=conv2,
num_channels=num_filters * 2,
reduction_ratio=reduction_ratio)
reduction_ratio=reduction_ratio,
name='fc' + name)
short = self.shortcut(input, num_filters * 2, stride)
short = self.shortcut(input, num_filters * 2, stride, name=name)
return fluid.layers.elementwise_add(x=short, y=scale, act='relu')
......@@ -151,7 +181,8 @@ class SE_ResNeXt():
filter_size,
stride=1,
groups=1,
act=None):
act=None,
name=None):
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -160,26 +191,42 @@ class SE_ResNeXt():
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
bias_attr=False)
return fluid.layers.batch_norm(input=conv, act=act)
bias_attr=False,
param_attr=ParamAttr(name=name + '_weights'), )
bn_name = name + "_bn"
return fluid.layers.batch_norm(
input=conv,
act=act,
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def squeeze_excitation(self, input, num_channels, reduction_ratio):
def squeeze_excitation(self,
input,
num_channels,
reduction_ratio,
name=None):
pool = fluid.layers.pool2d(
input=input, pool_size=0, pool_type='avg', global_pooling=True)
stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
squeeze = fluid.layers.fc(input=pool,
squeeze = fluid.layers.fc(
input=pool,
size=num_channels // reduction_ratio,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(
-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + '_sqz_weights'),
bias_attr=ParamAttr(name=name + '_sqz_offset'))
stdv = 1.0 / math.sqrt(squeeze.shape[1] * 1.0)
excitation = fluid.layers.fc(input=squeeze,
excitation = fluid.layers.fc(
input=squeeze,
size=num_channels,
act='sigmoid',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(
-stdv, stdv)))
initializer=fluid.initializer.Uniform(-stdv, stdv),
name=name + '_exc_weights'),
bias_attr=ParamAttr(name=name + '_exc_offset'))
scale = fluid.layers.elementwise_mul(x=input, y=excitation, axis=0)
return scale
......
fluid/PaddleCV/image_classification/models/shufflenet_v2.py
浏览文件 @ dc1b032d
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle.fluid as fluid
from paddle.fluid.initializer import MSRA
from paddle.fluid.param_attr import ParamAttr
__all__ = [
'ShuffleNetV2', 'ShuffleNetV2_x0_5', 'ShuffleNetV2_x1_0',
'ShuffleNetV2_x1_5', 'ShuffleNetV2_x2_0'
]
__all__ = ['ShuffleNetV2', 'ShuffleNetV2_x0_5_swish', 'ShuffleNetV2_x1_0_swish', 'ShuffleNetV2_x1_5_swish',
'ShuffleNetV2_x2_0_swish', 'ShuffleNetV2_x8_0_swish']
train_parameters = {
"input_size": [3, 224, 224],
......@@ -40,62 +35,43 @@ class ShuffleNetV2():
stage_out_channels = [-1, 24, 176, 352, 704, 1024]
elif scale == 2.0:
stage_out_channels = [-1, 24, 224, 488, 976, 2048]
elif scale == 8.0:
stage_out_channels = [-1, 48, 896, 1952, 3904, 8192]
else:
raise ValueError("""{} groups is not supported for
raise ValueError(
"""{} groups is not supported for
1x1 Grouped Convolutions""".format(num_groups))
#conv1
input_channel = stage_out_channels[1]
conv1 = self.conv_bn_layer(
input=input,
filter_size=3,
num_filters=input_channel,
padding=1,
stride=2)
pool1 = fluid.layers.pool2d(
input=conv1,
pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
conv1 = self.conv_bn_layer(input=input, filter_size=3, num_filters=input_channel, padding=1, stride=2,name='stage1_conv')
pool1 = fluid.layers.pool2d(input=conv1, pool_size=3, pool_stride=2, pool_padding=1, pool_type='max')
conv = pool1
# bottleneck sequences
for idxstage in range(len(stage_repeats)):
numrepeat = stage_repeats[idxstage]
output_channel = stage_out_channels[idxstage + 2]
output_channel = stage_out_channels[idxstage+2]
for i in range(numrepeat):
if i == 0:
conv = self.inverted_residual_unit(
input=conv,
num_filters=output_channel,
stride=2,
benchmodel=2)
conv = self.inverted_residual_unit(input=conv, num_filters=output_channel, stride=2,
benchmodel=2,name=str(idxstage+2)+'_'+str(i+1))
else:
conv = self.inverted_residual_unit(
input=conv,
num_filters=output_channel,
stride=1,
benchmodel=1)
conv = self.inverted_residual_unit(input=conv, num_filters=output_channel, stride=1,
benchmodel=1,name=str(idxstage+2)+'_'+str(i+1))
conv_last = self.conv_bn_layer(input=conv, filter_size=1, num_filters=stage_out_channels[-1],
padding=0, stride=1, name='conv5')
pool_last = fluid.layers.pool2d(input=conv_last, pool_size=7, pool_stride=1, pool_padding=0, pool_type='avg')
conv_last = self.conv_bn_layer(
input=conv,
filter_size=1,
num_filters=stage_out_channels[-1],
padding=0,
stride=1)
pool_last = fluid.layers.pool2d(
input=conv_last,
pool_size=7,
pool_stride=7,
pool_padding=0,
pool_type='avg')
output = fluid.layers.fc(input=pool_last,
size=class_dim,
param_attr=ParamAttr(initializer=MSRA()))
param_attr=ParamAttr(initializer=MSRA(),name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
return output
def conv_bn_layer(self,
input,
filter_size,
......@@ -104,7 +80,9 @@ class ShuffleNetV2():
padding,
num_groups=1,
use_cudnn=True,
if_act=True):
if_act=True,
name=None):
# print(num_groups)
conv = fluid.layers.conv2d(
input=input,
num_filters=num_filters,
......@@ -114,42 +92,53 @@ class ShuffleNetV2():
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(initializer=MSRA()),
param_attr=ParamAttr(initializer=MSRA(),name=name+'_weights'),
bias_attr=False)
out = int((input.shape[2] - 1)/float(stride) + 1)
# print(input.shape[1],(out, out), num_filters, (filter_size, filter_size), stride,
# (filter_size - 1) / 2, num_groups, name)
bn_name = name + '_bn'
if if_act:
return fluid.layers.batch_norm(input=conv, act='relu')
return fluid.layers.batch_norm(input=conv, act='swish',
param_attr = ParamAttr(name=bn_name+"_scale"),
bias_attr=ParamAttr(name=bn_name+"_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
else:
return fluid.layers.batch_norm(input=conv)
return fluid.layers.batch_norm(input=conv,
param_attr = ParamAttr(name=bn_name+"_scale"),
bias_attr=ParamAttr(name=bn_name+"_offset"),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def channel_shuffle(self, x, groups):
batchsize, num_channels, height, width = x.shape[0], x.shape[
1], x.shape[2], x.shape[3]
batchsize, num_channels, height, width = x.shape[0], x.shape[1], x.shape[2], x.shape[3]
channels_per_group = num_channels // groups
# reshape
x = fluid.layers.reshape(
x=x, shape=[batchsize, groups, channels_per_group, height, width])
x = fluid.layers.reshape(x=x, shape=[batchsize, groups, channels_per_group, height, width])
x = fluid.layers.transpose(x=x, perm=[0, 2, 1, 3, 4])
x = fluid.layers.transpose(x=x, perm=[0,2,1,3,4])
# flatten
x = fluid.layers.reshape(
x=x, shape=[batchsize, num_channels, height, width])
x = fluid.layers.reshape(x=x, shape=[batchsize, num_channels, height, width])
return x
def inverted_residual_unit(self, input, num_filters, stride, benchmodel):
def inverted_residual_unit(self, input, num_filters, stride, benchmodel, name=None):
assert stride in [1, 2], \
"supported stride are {} but your stride is {}".format([1,2], stride)
oup_inc = num_filters // 2
oup_inc = num_filters//2
inp = input.shape[1]
if benchmodel == 1:
x1, x2 = fluid.layers.split(
input,
num_or_sections=[input.shape[1] // 2, input.shape[1] // 2],
dim=1)
input, num_or_sections=[input.shape[1]//2, input.shape[1]//2], dim=1)
# x1 = input[:, :(input.shape[1]//2), :, :]
# x2 = input[:, (input.shape[1]//2):, :, :]
conv_pw = self.conv_bn_layer(
input=x2,
......@@ -158,7 +147,8 @@ class ShuffleNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name='stage_'+name+'_conv1')
conv_dw = self.conv_bn_layer(
input=conv_pw,
......@@ -167,7 +157,9 @@ class ShuffleNetV2():
stride=stride,
padding=1,
num_groups=oup_inc,
if_act=False)
if_act=False,
use_cudnn=False,
name='stage_'+name+'_conv2')
conv_linear = self.conv_bn_layer(
input=conv_dw,
......@@ -176,77 +168,88 @@ class ShuffleNetV2():
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name='stage_'+name+'_conv3')
out = fluid.layers.concat([x1, conv_linear], axis=1)
else:
#branch1
conv_dw = self.conv_bn_layer(
conv_dw_1 = self.conv_bn_layer(
input=input,
num_filters=inp,
filter_size=3,
stride=stride,
padding=1,
num_groups=inp,
if_act=False)
if_act=False,
use_cudnn=False,
name='stage_'+name+'_conv4')
conv_linear_1 = self.conv_bn_layer(
input=conv_dw,
input=conv_dw_1,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name='stage_'+name+'_conv5')
#branch2
conv_pw = self.conv_bn_layer(
conv_pw_2 = self.conv_bn_layer(
input=input,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name='stage_'+name+'_conv1')
conv_dw = self.conv_bn_layer(
input=conv_pw,
conv_dw_2 = self.conv_bn_layer(
input=conv_pw_2,
num_filters=oup_inc,
filter_size=3,
stride=stride,
padding=1,
num_groups=oup_inc,
if_act=False)
if_act=False,
use_cudnn=False,
name='stage_'+name+'_conv2')
conv_linear_2 = self.conv_bn_layer(
input=conv_dw,
input=conv_dw_2,
num_filters=oup_inc,
filter_size=1,
stride=1,
padding=0,
num_groups=1,
if_act=True)
if_act=True,
name='stage_'+name+'_conv3')
out = fluid.layers.concat([conv_linear_1, conv_linear_2], axis=1)
return self.channel_shuffle(out, 2)
def ShuffleNetV2_x0_5():
def ShuffleNetV2_x0_5_swish():
model = ShuffleNetV2(scale=0.5)
return model
def ShuffleNetV2_x1_0():
def ShuffleNetV2_x1_0_swish():
model = ShuffleNetV2(scale=1.0)
return model
def ShuffleNetV2_x1_5():
def ShuffleNetV2_x1_5_swish():
model = ShuffleNetV2(scale=1.5)
return model
def ShuffleNetV2_x2_0():
def ShuffleNetV2_x2_0_swish():
model = ShuffleNetV2(scale=2.0)
return model
def ShuffleNetV2_x8_0_swish():
model = ShuffleNetV2(scale=8.0)
return model
fluid/PaddleCV/image_classification/models/vgg.py
浏览文件 @ dc1b032d
......@@ -36,42 +36,37 @@ class VGGNet():
"supported layers are {} but input layer is {}".format(vgg_spec.keys(), layers)
nums = vgg_spec[layers]
conv1 = self.conv_block(input, 64, nums[0])
conv2 = self.conv_block(conv1, 128, nums[1])
conv3 = self.conv_block(conv2, 256, nums[2])
conv4 = self.conv_block(conv3, 512, nums[3])
conv5 = self.conv_block(conv4, 512, nums[4])
conv1 = self.conv_block(input, 64, nums[0], name="conv1_")
conv2 = self.conv_block(conv1, 128, nums[1], name="conv2_")
conv3 = self.conv_block(conv2, 256, nums[2], name="conv3_")
conv4 = self.conv_block(conv3, 512, nums[3], name="conv4_")
conv5 = self.conv_block(conv4, 512, nums[4], name="conv5_")
fc_dim = 4096
fc_name = ["fc6", "fc7", "fc8"]
fc1 = fluid.layers.fc(
input=conv5,
size=fc_dim,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
param_attr=fluid.param_attr.ParamAttr(name=fc_name[0] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[0] + "_offset"))
fc1 = fluid.layers.dropout(x=fc1, dropout_prob=0.5)
fc2 = fluid.layers.fc(
input=fc1,
size=fc_dim,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
param_attr=fluid.param_attr.ParamAttr(name=fc_name[1] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[1] + "_offset"))
fc2 = fluid.layers.dropout(x=fc2, dropout_prob=0.5)
out = fluid.layers.fc(
input=fc2,
size=class_dim,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.005)),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.1)))
param_attr=fluid.param_attr.ParamAttr(name=fc_name[2] + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(name=fc_name[2] + "_offset"))
return out
def conv_block(self, input, num_filter, groups):
def conv_block(self, input, num_filter, groups, name=None):
conv = input
for i in range(groups):
conv = fluid.layers.conv2d(
......@@ -82,9 +77,9 @@ class VGGNet():
padding=1,
act='relu',
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Normal(scale=0.01)),
name=name + str(i + 1) + "_weights"),
bias_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Constant(value=0.0)))
name=name + str(i + 1) + "_offset"))
return fluid.layers.pool2d(
input=conv, pool_size=2, pool_type='max', pool_stride=2)
......
fluid/PaddleCV/image_classification/reader.py
浏览文件 @ dc1b032d
......@@ -156,13 +156,12 @@ def _reader_creator(file_list,
for line in lines:
if mode == 'train' or mode == 'val':
img_path, label = line.split()
#img_path = img_path.replace("JPEG", "jpeg")
img_path = os.path.join(data_dir, img_path)
yield img_path, int(label)
elif mode == 'test':
img_path, label = line.split()
#img_path = img_path.replace("JPEG", "jpeg")
img_path = os.path.join(data_dir, img_path)
yield [img_path]
mapper = functools.partial(
......@@ -185,9 +184,11 @@ def train(data_dir=DATA_DIR, pass_id_as_seed=0):
def val(data_dir=DATA_DIR):
file_list = os.path.join(data_dir, 'val_list.txt')
return _reader_creator(file_list, 'val', shuffle=False, data_dir=data_dir)
return _reader_creator(file_list, 'val', shuffle=False,
data_dir=data_dir)
def test(data_dir=DATA_DIR):
file_list = os.path.join(data_dir, 'val_list.txt')
return _reader_creator(file_list, 'test', shuffle=False, data_dir=data_dir)
return _reader_creator(file_list, 'test', shuffle=False,
data_dir=data_dir)
fluid/PaddleCV/image_classification/reader_cv2.py
浏览文件 @ dc1b032d
......@@ -16,7 +16,6 @@ THREAD = 8
BUF_SIZE = 102400
DATA_DIR = 'data/ILSVRC2012'
img_mean = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))
img_std = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))
......@@ -40,8 +39,9 @@ def random_crop(img, size, scale=None, ratio=None):
w = 1. * aspect_ratio
h = 1. / aspect_ratio
bound = min((float(img.shape[1]) / img.shape[0]) / (w**2),
(float(img.shape[0]) / img.shape[1]) / (h**2))
bound = min((float(img.shape[0]) / img.shape[1]) / (w**2),
(float(img.shape[1]) / img.shape[0]) / (h**2))
scale_max = min(scale[1], bound)
scale_min = min(scale[0], bound)
......@@ -50,15 +50,14 @@ def random_crop(img, size, scale=None, ratio=None):
target_size = math.sqrt(target_area)
w = int(target_size * w)
h = int(target_size * h)
i = np.random.randint(0, img.shape[0] - w + 1)
j = np.random.randint(0, img.shape[1] - h + 1)
i = np.random.randint(0, img.size[0] - w + 1)
j = np.random.randint(0, img.size[1] - h + 1)
img = img[i:i + w, j:j + h, :]
img = img[i:i + h, j:j + w, :]
resized = cv2.resize(img, (size, size))
resized = cv2.resize(img, (size, size), interpolation=cv2.INTER_LANCZOS4)
return resized
def distort_color(img):
return img
......@@ -68,7 +67,7 @@ def resize_short(img, target_size):
percent = float(target_size) / min(img.shape[0], img.shape[1])
resized_width = int(round(img.shape[1] * percent))
resized_height = int(round(img.shape[0] * percent))
resized = cv2.resize(img, (resized_width, resized_height))
resized = cv2.resize(img, (resized_width, resized_height), interpolation=cv2.INTER_LANCZOS4)
return resized
......@@ -140,16 +139,19 @@ def _reader_creator(file_list,
shuffle=False,
color_jitter=False,
rotate=False,
data_dir=DATA_DIR):
data_dir=DATA_DIR,
pass_id_as_seed=0):
def reader():
with open(file_list) as flist:
full_lines = [line.strip() for line in flist]
if shuffle:
np.random.shuffle(lines)
if pass_id_as_seed:
np.random.seed(pass_id_as_seed)
np.random.shuffle(full_lines)
if mode == 'train' and os.getenv('PADDLE_TRAINING_ROLE'):
# distributed mode if the env var `PADDLE_TRAINING_ROLE` exits
trainer_id = int(os.getenv("PADDLE_TRAINER_ID", "0"))
trainer_count = int(os.getenv("PADDLE_TRAINERS", "1"))
trainer_count = int(os.getenv("PADDLE_TRAINERS_NUM", "1"))
per_node_lines = len(full_lines) // trainer_count
lines = full_lines[trainer_id * per_node_lines:(trainer_id + 1)
* per_node_lines]
......@@ -159,6 +161,7 @@ def _reader_creator(file_list,
len(full_lines)))
else:
lines = full_lines
for line in lines:
if mode == 'train' or mode == 'val':
img_path, label = line.split()
......@@ -166,21 +169,25 @@ def _reader_creator(file_list,
img_path = os.path.join(data_dir, img_path)
yield img_path, int(label)
elif mode == 'test':
img_path = os.path.join(DATA_DIR, line)
img_path, label = line.split()
img_path = img_path.replace("JPEG", "jpeg")
img_path = os.path.join(data_dir, img_path)
yield [img_path]
image_mapper = functools.partial(
process_image,
mode=mode,
color_jitter=color_jitter,
rotate=color_jitter,
rotate=rotate,
crop_size=224)
reader = paddle.reader.xmap_readers(
image_mapper, reader, THREAD, BUF_SIZE, order=False)
return reader
def train(data_dir=DATA_DIR):
def train(data_dir=DATA_DIR, pass_id_as_seed=0):
file_list = os.path.join(data_dir, 'train_list.txt')
return _reader_creator(
file_list,
......@@ -188,14 +195,17 @@ def train(data_dir=DATA_DIR):
shuffle=True,
color_jitter=False,
rotate=False,
data_dir=data_dir)
data_dir=data_dir,
pass_id_as_seed=pass_id_as_seed)
def val(data_dir=DATA_DIR):
file_list = os.path.join(data_dir, 'val_list.txt')
return _reader_creator(file_list, 'val', shuffle=False, data_dir=data_dir)
return _reader_creator(file_list, 'val', shuffle=False,
data_dir=data_dir)
def test(data_dir=DATA_DIR):
file_list = os.path.join(data_dir, 'val_list.txt')
return _reader_creator(file_list, 'test', shuffle=False, data_dir=data_dir)
return _reader_creator(file_list, 'test', shuffle=False,
data_dir=data_dir)
fluid/PaddleCV/image_classification/run.sh
浏览文件 @ dc1b032d
......@@ -12,7 +12,6 @@ python train.py \
--lr=0.1 \
--num_epochs=200 \
--l2_decay=1.2e-4 \
--model_category=models_name \
# >log_SE_ResNeXt50_32x4d.txt 2>&1 &
#AlexNet:
#python train.py \
......@@ -23,7 +22,6 @@ python train.py \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --model_category=models_name \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.01 \
......@@ -38,7 +36,6 @@ python train.py \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --model_category=models_name \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.1 \
......@@ -51,12 +48,63 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --model_category=models_name \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
# --l2_decay=4e-5
#ResNet18:
#python train.py \
# --model=ResNet18 \
# --batch_size=256 \
# --total_images=1281167 \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --l2_decay=1e-4
#ResNet34:
#python train.py \
# --model=ResNet34 \
# --batch_size=256 \
# --total_images=1281167 \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --l2_decay=1e-4
#ShuffleNetv2:
#python train.py \
# --model=ShuffleNetV2 \
# --batch_size=1024 \
# --total_images=1281167 \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay_with_warmup \
# --lr=0.5 \
# --num_epochs=240 \
# --l2_decay=4e-5
#GoogleNet:
#python train.py \
# --model=GoogleNet \
# --batch_size=256 \
# --total_images=1281167 \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.01 \
# --num_epochs=200 \
# --l2_decay=1e-4
#ResNet50:
#python train.py \
# --model=ResNet50 \
......@@ -66,7 +114,6 @@ python train.py \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --model_category=models_name \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.1 \
......@@ -80,7 +127,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --model_category=models_name \
# --with_mem_opt=True \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
......@@ -96,7 +142,6 @@ python train.py \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --lr_strategy=piecewise_decay \
# --model_category=models_name \
# --with_mem_opt=True \
# --lr=0.1 \
# --num_epochs=120 \
......@@ -111,7 +156,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --lr_strategy=cosine_decay \
# --model_category=models_name \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=200 \
......@@ -126,7 +170,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --lr_strategy=cosine_decay \
# --model_category=models_name \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=200 \
......@@ -141,7 +184,6 @@ python train.py \
# --image_shape=3,224,224 \
# --lr_strategy=cosine_decay \
# --class_dim=1000 \
# --model_category=models_name \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=90 \
......@@ -158,7 +200,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.01 \
# --num_epochs=90 \
# --model_category=models_name \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --l2_decay=3e-4
......@@ -171,7 +212,6 @@ python train.py \
# --class_dim=1000 \
# --lr_strategy=cosine_decay \
# --image_shape=3,224,224 \
# --model_category=models_name \
# --model_save_dir=output/ \
# --lr=0.01 \
# --num_epochs=90 \
......@@ -189,7 +229,6 @@ python train.py \
# --lr=0.01 \
# --num_epochs=90 \
# --with_mem_opt=True \
# --model_category=models_name \
# --model_save_dir=output/ \
# --l2_decay=3e-4
......@@ -204,7 +243,6 @@ python train.py \
# --lr=0.001 \
# --num_epochs=120 \
# --with_mem_opt=False \
# --model_category=models_name \
# --model_save_dir=output/ \
# --lr_strategy=adam \
# --use_gpu=False
......
fluid/PaddleCV/image_classification/train.py
浏览文件 @ dc1b032d
......@@ -10,14 +10,15 @@ import math
import paddle
import paddle.fluid as fluid
import paddle.dataset.flowers as flowers
import reader
import reader as reader
import argparse
import functools
import subprocess
import utils
from utils.learning_rate import cosine_decay
import models
from utils.fp16_utils import create_master_params_grads, master_param_to_train_param
from utility import add_arguments, print_arguments
from utils.utility import add_arguments, print_arguments
from utils.learning_rate import cosine_decay_with_warmup
IMAGENET1000 = 1281167
......@@ -44,19 +45,6 @@ add_arg('fp16', bool, False, "Enable half precision
add_arg('scale_loss', float, 1.0, "Scale loss for fp16." )
add_arg('l2_decay', float, 1e-4, "L2_decay parameter.")
add_arg('momentum_rate', float, 0.9, "momentum_rate.")
# yapf: enable
def set_models(model_category):
global models
assert model_category in ["models", "models_name"
], "{} is not in lists: {}".format(
model_category, ["models", "models_name"])
if model_category == "models_name":
import models_name as models
else:
import models as models
def optimizer_setting(params):
ls = params["learning_strategy"]
......@@ -68,8 +56,7 @@ def optimizer_setting(params):
else:
total_images = params["total_images"]
batch_size = ls["batch_size"]
step = int(total_images / batch_size + 1)
step = int(math.ceil(float(total_images) / batch_size))
bd = [step * e for e in ls["epochs"]]
base_lr = params["lr"]
lr = []
......@@ -88,16 +75,34 @@ def optimizer_setting(params):
batch_size = ls["batch_size"]
l2_decay = params["l2_decay"]
momentum_rate = params["momentum_rate"]
step = int(total_images / batch_size + 1)
step = int(math.ceil(float(total_images) / batch_size))
lr = params["lr"]
num_epochs = params["num_epochs"]
optimizer = fluid.optimizer.Momentum(
learning_rate=fluid.layers.cosine_decay(
learning_rate=lr, step_each_epoch=step, epochs=num_epochs),
momentum=momentum_rate,
regularization=fluid.regularizer.L2Decay(l2_decay))
elif ls["name"] == "cosine_warmup_decay":
if "total_images" not in params:
total_images = IMAGENET1000
else:
total_images = params["total_images"]
batch_size = ls["batch_size"]
l2_decay = params["l2_decay"]
momentum_rate = params["momentum_rate"]
step = int(math.ceil(float(total_images) / batch_size))
lr = params["lr"]
num_epochs = params["num_epochs"]
optimizer = fluid.optimizer.Momentum(
learning_rate=cosine_decay(
learning_rate=cosine_decay_with_warmup(
learning_rate=lr, step_each_epoch=step, epochs=num_epochs),
momentum=momentum_rate,
regularization=fluid.regularizer.L2Decay(l2_decay))
elif ls["name"] == "linear_decay":
if "total_images" not in params:
total_images = IMAGENET1000
......@@ -119,6 +124,25 @@ def optimizer_setting(params):
elif ls["name"] == "adam":
lr = params["lr"]
optimizer = fluid.optimizer.Adam(learning_rate=lr)
elif ls["name"] == "rmsprop_cosine":
if "total_images" not in params:
total_images = IMAGENET1000
else:
total_images = params["total_images"]
batch_size = ls["batch_size"]
l2_decay = params["l2_decay"]
momentum_rate = params["momentum_rate"]
step = int(math.ceil(float(total_images) / batch_size))
lr = params["lr"]
num_epochs = params["num_epochs"]
optimizer = fluid.optimizer.RMSProp(
learning_rate=fluid.layers.cosine_decay(
learning_rate=lr, step_each_epoch=step, epochs=num_epochs),
momentum=momentum_rate,
regularization=fluid.regularizer.L2Decay(l2_decay),
# RMSProp Optimizer: Apply epsilon=1 on ImageNet.
epsilon=1
)
else:
lr = params["lr"]
l2_decay = params["l2_decay"]
......@@ -130,7 +154,6 @@ def optimizer_setting(params):
return optimizer
def net_config(image, label, model, args):
model_list = [m for m in dir(models) if "__" not in m]
assert args.model in model_list, "{} is not lists: {}".format(args.model,
......@@ -220,6 +243,13 @@ def build_program(is_train, main_prog, startup_prog, args):
else:
return py_reader, avg_cost, acc_top1, acc_top5
def get_device_num():
visible_device = os.getenv('CUDA_VISIBLE_DEVICES')
if visible_device:
device_num = len(visible_device.split(','))
else:
device_num = subprocess.check_output(['nvidia-smi','-L']).decode().count('\n')
return device_num
def train(args):
# parameters from arguments
......@@ -268,12 +298,7 @@ def train(args):
exe, pretrained_model, main_program=train_prog, predicate=if_exist)
if args.use_gpu:
visible_device = os.getenv('CUDA_VISIBLE_DEVICES')
if visible_device:
device_num = len(visible_device.split(','))
else:
device_num = subprocess.check_output(
['nvidia-smi', '-L']).decode().count('\n')
device_num = get_device_num()
else:
device_num = 1
train_batch_size = args.batch_size / device_num
......@@ -345,8 +370,8 @@ def train(args):
if batch_id % 10 == 0:
print("Pass {0}, trainbatch {1}, loss {2}, \
acc1 {3}, acc5 {4}, lr{5}, time {6}"
.format(pass_id, batch_id, loss, acc1, acc5, "%.5f" %
acc1 {3}, acc5 {4}, lr {5}, time {6}"
.format(pass_id, batch_id, "%.5f"%loss, "%.5f"%acc1, "%.5f"%acc5, "%.5f" %
lr, "%2.2f sec" % period))
sys.stdout.flush()
batch_id += 1
......@@ -378,7 +403,7 @@ def train(args):
if test_batch_id % 10 == 0:
print("Pass {0},testbatch {1},loss {2}, \
acc1 {3},acc5 {4},time {5}"
.format(pass_id, test_batch_id, loss, acc1, acc5,
.format(pass_id, test_batch_id, "%.5f"%loss,"%.5f"%acc1, "%.5f"%acc5,
"%2.2f sec" % period))
sys.stdout.flush()
test_batch_id += 1
......@@ -391,8 +416,8 @@ def train(args):
print("End pass {0}, train_loss {1}, train_acc1 {2}, train_acc5 {3}, "
"test_loss {4}, test_acc1 {5}, test_acc5 {6}".format(
pass_id, train_loss, train_acc1, train_acc5, test_loss,
test_acc1, test_acc5))
pass_id, "%.5f"%train_loss, "%.5f"%train_acc1, "%.5f"%train_acc5, "%.5f"%test_loss,
"%.5f"%test_acc1, "%.5f"%test_acc5))
sys.stdout.flush()
model_path = os.path.join(model_save_dir + '/' + model_name,
......@@ -429,7 +454,6 @@ def train(args):
def main():
args = parser.parse_args()
set_models(args.model_category)
print_arguments(args)
train(args)
......
fluid/PaddleCV/image_classification/utils/__init__.py
浏览文件 @ dc1b032d
from .learning_rate import cosine_decay, lr_warmup
from .fp16_utils import create_master_params_grads, master_param_to_train_param
from .utility import add_arguments, print_arguments
fluid/PaddleCV/image_classification/utils/learning_rate.py
浏览文件 @ dc1b032d
......@@ -21,6 +21,33 @@ def cosine_decay(learning_rate, step_each_epoch, epochs=120):
(ops.cos(epoch * (math.pi / epochs)) + 1)/2
return decayed_lr
def cosine_decay_with_warmup(learning_rate, step_each_epoch, epochs=120):
"""Applies cosine decay to the learning rate.
lr = 0.05 * (math.cos(epoch * (math.pi / 120)) + 1)
decrease lr for every mini-batch and start with warmup.
"""
global_step = _decay_step_counter()
lr = fluid.layers.tensor.create_global_var(
shape=[1],
value=0.0,
dtype='float32',
persistable=True,
name="learning_rate")
warmup_epoch = fluid.layers.fill_constant(
shape=[1], dtype='float32', value=float(5), force_cpu=True)
with init_on_cpu():
epoch = ops.floor(global_step / step_each_epoch)
with control_flow.Switch() as switch:
with switch.case(epoch < warmup_epoch):
decayed_lr = learning_rate * (global_step / (step_each_epoch * warmup_epoch))
fluid.layers.tensor.assign(input=decayed_lr, output=lr)
with switch.default():
decayed_lr = learning_rate * \
(ops.cos((global_step - warmup_epoch * step_each_epoch) * (math.pi / (epochs * step_each_epoch))) + 1)/2
fluid.layers.tensor.assign(input=decayed_lr, output=lr)
return lr
def lr_warmup(learning_rate, warmup_steps, start_lr, end_lr):
""" Applies linear learning rate warmup for distributed training
......
fluid/PaddleCV/image_classification/utility.py → fluid/PaddleCV/image_classification/utils/utility.py
浏览文件 @ dc1b032d
......@@ -37,10 +37,10 @@ def print_arguments(args):
:param args: Input argparse.Namespace for printing.
:type args: argparse.Namespace
"""
print("----------- Configuration Arguments -----------")
print("------------- Configuration Arguments -------------")
for arg, value in sorted(six.iteritems(vars(args))):
print("%s: %s" % (arg, value))
print("------------------------------------------------")
print("%25s : %s" % (arg, value))
print("----------------------------------------------------")
def add_arguments(argname, type, default, help, argparser, **kwargs):
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或