未验证 提交 843bdbcf 编写于 作者: C cuicheng01 提交者: GitHub

Merge branch 'PaddlePaddle:develop' into develop

......@@ -8,30 +8,40 @@
<div align="center">
<img src="./docs/images/class_simple.gif" width = "600" />
PULC实用图像分类模型效果展示
<p>PULC实用图像分类模型效果展示</p>
</div>
&nbsp;
<div align="center">
<img src="./docs/images/recognition.gif" width = "400" />
PP-ShiTu图像识别系统效果展示
<p>PP-ShiTu图像识别系统效果展示</p>
</div>
## 近期更新
- 📢将于**<u>6月15-6月17日晚20:30</u>**进行为期三天的课程直播,详细介绍超轻量图像分类方案,对各场景模型优化原理及使用方式进行拆解,之后还有产业案例全流程实操,对各类痛难点解决方案进行手把手教学,加上现场互动答疑,抓紧扫码上车吧!
<div align="center">
<img src="https://user-images.githubusercontent.com/80816848/173404459-9426c0ed-4801-4f75-876f-2e6ec47255f5.png" width = "200" height = "200"/>
</div>
- 🔥️ 2022.6.15 发布[PULC超轻量图像分类实用方案](docs/zh_CN/PULC/PULC_train.md),CPU推理3ms,精度比肩SwinTransformer,覆盖人、车、OCR场景九大常见任务。
- 2022.5.26 [飞桨产业实践范例直播课](http://aglc.cn/v-c4FAR),解读**超轻量重点区域人员出入管理方案**
- 2022.5.23 新增[人员出入管理范例库](https://aistudio.baidu.com/aistudio/projectdetail/4094475),具体内容可以在 AI Stuio 上体验。
- 2022.5.20 上线[PP-HGNet](./docs/zh_CN/models/PP-HGNet.md), [PP-LCNetv2](./docs/zh_CN/models/PP-LCNetV2.md)
- 2022.4.21 新增 CVPR2022 oral论文 [MixFormer](https://arxiv.org/pdf/2204.02557.pdf) 相关[代码](https://github.com/PaddlePaddle/PaddleClas/pull/1820/files)
- [more](./docs/zh_CN/others/update_history.md)
## 特性
支持多种图像分类、识别相关算法,在此基础上打造[PULC超轻量图像分类方案](docs/zh_CN/PULC/PULC_person_exists.md)[PP-ShiTu图像识别系统](./docs/zh_CN/quick_start/quick_start_recognition.md)
PaddleClas发布了[PP-HGNet](docs/zh_CN/models/PP-HGNet.md)[PP-LCNetv2](docs/zh_CN/models/PP-LCNetV2.md)[PP-LCNet](docs/zh_CN/models/PP-LCNet.md)[SSLD半监督知识蒸馏方案](docs/zh_CN/advanced_tutorials/ssld.md)等算法,
并支持多种图像分类、识别相关算法,在此基础上打造[PULC超轻量图像分类方案](docs/zh_CN/PULC/PULC_quickstart.md)[PP-ShiTu图像识别系统](./docs/zh_CN/quick_start/quick_start_recognition.md)
![](https://user-images.githubusercontent.com/19523330/173273046-239a42da-c88d-4c2c-94b1-2134557afa49.png)
......
......@@ -4,10 +4,24 @@
## Introduction
PaddleClas is an image recognition toolset for industry and academia, helping users train better computer vision models and apply them in real scenarios.
PaddleClas is an image classification and image recognition toolset for industry and academia, helping users train better computer vision models and apply them in real scenarios.
**Recent updates**
<div align="center">
<img src="./docs/images/class_simple.gif" width = "600" />
PULC demo images
</div>
&nbsp;
<div align="center">
<img src="./docs/images/recognition.gif" width = "400" />
PP-ShiTu demo images
</div>
**Recent updates**
- 2022.6.15 Release [**P**ractical **U**ltra **L**ight-weight image **C**lassification solutions](./docs/en/PULC/PULC_quickstart_en.md). PULC models inference within 3ms on CPU devices, with accuracy comparable with SwinTransformer. We also release 9 practical models covering pedestrian, vehicle and OCR.
- 2022.4.21 Added the related [code](https://github.com/PaddlePaddle/PaddleClas/pull/1820/files) of the CVPR2022 oral paper [MixFormer](https://arxiv.org/pdf/2204.02557.pdf).
- 2021.09.17 Add PP-LCNet series model developed by PaddleClas, these models show strong competitiveness on Intel CPUs.
......@@ -19,24 +33,12 @@ For the introduction of PP-LCNet, please refer to [paper](https://arxiv.org/pdf/
## Features
- A practical image recognition system consist of detection, feature learning and retrieval modules, widely applicable to all types of image recognition tasks.
Four sample solutions are provided, including product recognition, vehicle recognition, logo recognition and animation character recognition.
- Rich library of pre-trained models: Provide a total of 164 ImageNet pre-trained models in 35 series, among which 6 selected series of models support fast structural modification.
- Comprehensive and easy-to-use feature learning components: 12 metric learning methods are integrated and can be combined and switched at will through configuration files.
- SSLD knowledge distillation: The 14 classification pre-training models generally improved their accuracy by more than 3%; among them, the ResNet50_vd model achieved a Top-1 accuracy of 84.0% on the Image-Net-1k dataset and the Res2Net200_vd pre-training model achieved a Top-1 accuracy of 85.1%.
PaddleClas release PP-HGNet、PP-LCNetv2、 PP-LCNet and **S**imple **S**emi-supervised **L**abel **D**istillation algorithms, and support plenty of
image classification and image recognition algorithms.
Based on th algorithms above, PaddleClas release PP-ShiTu image recognition system and [**P**ractical **U**ltra **L**ight-weight image **C**lassification solutions](docs/en/PULC/PULC_quickstart_en.md).
- Data augmentation: Provide 8 data augmentation algorithms such as AutoAugment, Cutout, Cutmix, etc. with detailed introduction, code replication and evaluation of effectiveness in a unified experimental environment.
<div align="center">
<img src="./docs/images/recognition_en.gif" width = "400" />
</div>
![](https://user-images.githubusercontent.com/19523330/173347904-f2998e00-7b86-4adf-b546-23c684fc67b9.png)
## Welcome to Join the Technical Exchange Group
......@@ -48,11 +50,13 @@ Four sample solutions are provided, including product recognition, vehicle recog
</div>
## Quick Start
Quick experience of image recognition:[Link](./docs/en/tutorials/quick_start_recognition_en.md)
Quick experience of PP-ShiTu image recognition system:[Link](./docs/en/tutorials/quick_start_recognition_en.md)
Quick experience of **P**ractical **U**ltra **L**ight-weight image **C**lassification models:[Link](docs/en/PULC/PULC_quickstart.md)
## Tutorials
- [Quick Installation](./docs/en/tutorials/install_en.md)
- [Practical Ultra Light-weight image Classification solutions](./docs/en/PULC/PULC_quickstart_en.md)
- [Quick Start of Recognition](./docs/en/tutorials/quick_start_recognition_en.md)
- [Introduction to Image Recognition Systems](#Introduction_to_Image_Recognition_Systems)
- [Demo images](#Demo_images)
......@@ -83,6 +87,14 @@ Quick experience of image recognition:[Link](./docs/en/tutorials/quick_start_r
- [License](#License)
- [Contribution](#Contribution)
<a name="Introduction_to_PULC"></a>
## Introduction to Practical Ultra Light-weight image Classification solutions
<div align="center">
<img src="https://user-images.githubusercontent.com/19523330/173011854-b10fcd7a-b799-4dfd-a1cf-9504952a3c44.png" width = "800" />
</div>
PULC solutions consists of PP-LCNet light-weight backbone, SSLD pretrained models, Ensemble of Data Augmentation strategy and SKL-UGI knowledge distillation.
PULC models inference within 3ms on CPU devices, with accuracy comparable with SwinTransformer. We also release 9 practical models covering pedestrian, vehicle and OCR.
<a name="Introduction_to_Image_Recognition_Systems"></a>
## Introduction to Image Recognition Systems
......@@ -97,8 +109,13 @@ Image recognition can be divided into three steps:
For a new unknown category, there is no need to retrain the model, just prepare images of new category, extract features and update retrieval database and the category can be recognised.
<a name="Demo_images"></a>
## Demo images [more](https://github.com/PaddlePaddle/PaddleClas/tree/release/2.2/docs/images/recognition/more_demo_images)
## PULC demo images
<div align="center">
<img src="docs/images/classification.gif">
</div>
<a name="Rec_Demo_images"></a>
## Image Recognition Demo images [more](https://github.com/PaddlePaddle/PaddleClas/tree/release/2.2/docs/images/recognition/more_demo_images)
- Product recognition
<div align="center">
<img src="https://user-images.githubusercontent.com/18028216/122769644-51604f80-d2d7-11eb-8290-c53b12a5c1f6.gif" width = "400" />
......
......@@ -30,7 +30,7 @@ PostProcess:
main_indicator: ThreshOutput
ThreshOutput:
threshold: 0.5
label_0: nocar
label_0: no_car
label_1: contains_car
SavePreLabel:
save_dir: ./pre_label/
Global:
infer_imgs: "./images/PULC/car_exists/objects365_00001507.jpeg"
inference_model_dir: "./models/car_exists_infer"
batch_size: 1
use_gpu: True
enable_mkldnn: False
cpu_num_threads: 10
enable_benchmark: True
use_fp16: False
ir_optim: True
use_tensorrt: False
gpu_mem: 8000
enable_profile: False
PreProcess:
transform_ops:
- ResizeImage:
resize_short: 256
- CropImage:
size: 224
- NormalizeImage:
scale: 0.00392157
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
channel_num: 3
- ToCHWImage:
PostProcess:
main_indicator: ThreshOutput
ThreshOutput:
threshold: 0.5
label_0: nocar
label_1: contains_car
SavePreLabel:
save_dir: ./pre_label/
Global:
infer_imgs: "./images/PULC/vehicle_exists/objects365_00001507.jpeg"
inference_model_dir: "./models/vehicle_exists_infer"
batch_size: 1
use_gpu: True
enable_mkldnn: False
cpu_num_threads: 10
enable_benchmark: True
use_fp16: False
ir_optim: True
use_tensorrt: False
gpu_mem: 8000
enable_profile: False
PreProcess:
transform_ops:
- ResizeImage:
resize_short: 256
- CropImage:
size: 224
- NormalizeImage:
scale: 0.00392157
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
channel_num: 3
- ToCHWImage:
PostProcess:
main_indicator: ThreshOutput
ThreshOutput:
threshold: 0.5
label_0: no_vehicle
label_1: contains_vehicle
SavePreLabel:
save_dir: ./pre_label/
# PULC Model Zoo
------
The PULC model zoo is provided here, mainly providing indicators, model storage size, and download links of the model. The pre-trained model can be used for fine-tuning training, and the inference model can be directly used for prediction and deployment.
|Model name| Model Description | Metrics |Storage Size| Latency| Download Address|
| --- | --- | --- | --- | --- | --- |
| person_exists |[Human Exists Classification](PULC_person_exists_en.md)| 95.60 |6.5M|2.58ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/person_exists_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/person_exists_pretrained.pdparams)|
| person_attribute |[Pedestrian Attribute Classification](PULC_person_attribute_en.md)| 78.59 |6.6M|2.01ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/person_attribute_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/person_attribute_pretrained.pdparams)|
| safety_helmet |[Classification of Wheather Wearing Safety Helmet](PULC_safety_helmet_en.md)| 99.38 |6.5M|2.03ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/safety_helmet_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/safety_helmet_pretrained.pdparams)|
| traffic_sign |[Traffic Sign Classification](PULC_traffic_sign_en.md)| 98.35 |8.2M|2.10ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/traffic_sign_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/traffic_sign_pretrained.pdparams)|
| vehicle_attribute |[Vehicle Attribute Classification](PULC_vehicle_attribute_en.md)| 90.81 |7.2M|2.36ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/vehicle_attribute_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/vehicle_attribute_pretrained.pdparams)|
| car_exists |[Car Exists Classification](PULC_car_exists_en.md) | 95.92 | 6.6M | 2.38ms |[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/car_exists_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/car_exists_pretrained.pdparams)|
| text_image_orientation |[Text Image Orientation Classification](PULC_text_image_orientation_en.md)| 99.06 | 6.5M | 2.16ms |[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/text_image_orientation_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/text_image_orientation_pretrained.pdparams)|
| textline_orientation |[Text-line Orientation Classification](PULC_textline_orientation_en.md)| 96.01 |6.5M|2.72ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/textline_orientation_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/textline_orientation_pretrained.pdparams)|
| language_classification |[Language Classification](PULC_language_classification_en.md)| 99.26 |6.5M|2.58ms|[inference model](https://paddleclas.bj.bcebos.com/models/PULC/inference/language_classification_infer.tar) / [pretrained model](https://paddleclas.bj.bcebos.com/models/PULC/pretrained/language_classification_pretrained.pdparams)|
**Note:**
* The backbone of all the above models is PPLCNet_x1_0. The different sizes of some models are caused by the different output sizes of the classification layer. The inference time is tested on the Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz. During the test process, the MKLDNN acceleration strategy is turned on, and the number of threads is 10. There will be slight fluctuations during the speed test process.
* The evaluation indicators of person_exists, safety_helmet, and car_exists are TprAtFpr. The evaluation indicators of person_attribute and vehicle_attribute are ma. The evaluation indicators of traffic_sign, text_image_orientation, textline_orientation and language_classification are Top-1 Acc.
# PULC Quick Start
------
This document introduces the prediction using PULC series model based on PaddleClas wheel.
## Catalogue
- [1. Installation](#1)
- [1.1 PaddlePaddle Installation](#11)
- [1.2 PaddleClas wheel Installation](#12)
- [2. Quick Start](#2)
- [2.1 Predicion with Command Line](#2.1)
- [2.2 Predicion with Python](#2.2)
- [2.3 Supported Model List](#2.3)
- [3. Summary](#3)
<a name="1"></a>
## 1. Installation
<a name="1.1"></a>
### 1.1 PaddlePaddle Installation
- Run the following command to install if CUDA9 or CUDA10 is available.
```bash
python3 -m pip install paddlepaddle-gpu -i https://mirror.baidu.com/pypi/simple
```
- Run the following command to install if GPU device is unavailable.
```bash
python3 -m pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple
```
Please refer to [PaddlePaddle Installation](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/en/install/pip/linux-pip_en.html) for more information about installation, for examples other versions.
<a name="1.2"></a>
### 1.2 PaddleClas wheel Installation
```bash
pip3 install paddleclas
```
<a name="2"></a>
## 2. Quick Start
PaddleClas provides a series of test cases, which contain demos of different scenes about people, cars, OCR, etc. Click [here](https://paddleclas.bj.bcebos.com/data/PULC/pulc_demo_imgs.zip) to download the data.
<a name="2.1"></a>
### 2.1 Predicion with Command Line
```
cd /path/to/pulc_demo_imgs
```
The prediction command:
```bash
paddleclas --model_name=person_exists --infer_imgs=pulc_demo_imgs/person_exists/objects365_01780782.jpg
```
Result:
```
>>> result
class_ids: [0], scores: [0.9955421453341842], label_names: ['nobody'], filename: pulc_demo_imgs/person_exists/objects365_01780782.jpg
Predict complete!
```
`Nobody` means there is no one in the image, `someone` means there is someone in the image. Therefore, the prediction result indicates that there is no one in the figure.
**Note**: The "--infer_imgs" argument specify the image(s) to be predict, and you can also specify a directoy contains images. If use other model, you can specify the `--model_name` argument. Please refer to [2.3 Supported Model List](#2.3) for the supported model list.
<a name="2.2"></a>
### 2.2 Predicion with Python
You can also use in Python:
```python
import paddleclas
model = paddleclas.PaddleClas(model_name="person_exists")
result = model.predict(input_data="pulc_demo_imgs/person_exists/objects365_01780782.jpg")
print(next(result))
```
The printed result information:
```
>>> result
[{'class_ids': [0], 'scores': [0.9955421453341842], 'label_names': ['nobody'], 'filename': 'pulc_demo_imgs/person_exists/objects365_01780782.jpg'}]
```
**Note**: `model.predict()` is a generator, so `next()` or `for` is needed to call it. This would to predict by batch that length is `batch_size`, default by 1. You can specify the argument `batch_size` and `model_name` when instantiating PaddleClas object, for example: `model = paddleclas.PaddleClas(model_name="person_exists", batch_size=2)`. Please refer to [2.3 Supported Model List](#2.3) for the supported model list.
<a name="2.3"></a>
### 2.3 Supported Model List
The name of PULC series models are as follows:
| Name | Intro |
| --- | --- |
| person_exists | Human Exists Classification |
| person_attribute | Pedestrian Attribute Classification |
| safety_helmet | Classification of Wheather Wearing Safety Helmet |
| traffic_sign | Traffic Sign Classification |
| vehicle_attribute | Vehicle Attribute Classification |
| car_exists | Car Exists Classification |
| text_image_orientation | Text Image Orientation Classification |
| textline_orientation | Text-line Orientation Classification |
| language_classification | Language Classification |
<a name="3"></a>
## 3. Summary
The PULC series models have been verified to be effective in different scenarios about people, vehicles, OCR, etc. The ultra lightweight model can achieve the accuracy close to SwinTransformer model, and the speed is increased by 40+ times. And PULC also provides the whole process of dataset getting, model training, model compression and deployment. Please refer to [Human Exists Classification](PULC_person_exists_en.md)[Pedestrian Attribute Classification](PULC_person_attribute_en.md)[Classification of Wheather Wearing Safety Helmet](PULC_safety_helmet_en.md)[Traffic Sign Classification](PULC_traffic_sign_en.md)[Vehicle Attribute Classification](PULC_vehicle_attribute_en.md)[Car Exists Classification](PULC_car_exists_en.md)[Text Image Orientation Classification](PULC_text_image_orientation_en.md)[Text-line Orientation Classification](PULC_textline_orientation_en.md)[Language Classification](PULC_language_classification_en.md) for more information about different scenarios.
## Practical Ultra Lightweight Classification scheme PULC
------
## Catalogue
- [1. Introduction of PULC solution](#1)
- [2. Data preparation](#2)
- [2.1 Dataset format description](#2.1)
- [2.2 Annotation file generation method](#2.2)
- [3. Training with standard classification configuration](#3)
- [3.1 PP-LCNet as backbone](#3.1)
- [3.2 SSLD pretrained model](#3.2)
- [3.3 EDA strategy](#3.3)
- [3.4 SKL-UGI knowledge distillation](#3.4)
- [3.5 Summary](#3.5)
- [4. Hyperparameter Search](#4)
- [4.1 Search based on default configuration](#4.1)
- [4.2 Custom search configuration](#4.2)
<a name="1"></a>
### 1. Introduction of PULC solution
Image classification is one of the basic algorithms of computer vision, and it is also the most common algorithm in enterprise applications, and further, it is also an important part of many CV applications. In recent years, the backbone network model has developed rapidly, and the accuracy record of ImageNet has been continuously refreshed. However, the performance of these models in practical scenarios is sometimes unsatisfactory. On the one hand, models with high precision tend to have large storage and slow inference speed, which are often difficult to meet actual deployment requirements; on the other hand, after selecting a suitable model, experienced engineers are often required to adjust parameters, which is time-consuming and labor-intensive. In order to solve the problems of enterprise application and make the training and parameter adjustment of classification models easier, PaddleClas summarized and launched a Practical Ultra Lightweight Classification (PULC) solution. PULC integrates various state-of-the-art algorithms such as backbone network, data augmentation and distillation, etc., and finally can automatically obtain a lightweight and high-precision image classification model.
The PULC solution has been verified to be effective in many scenarios, such as human-related scenarios, car-related scenarios, and OCR-related scenarios. With an ultra-lightweight model, the accuracy close to SwinTransformer can be achieved, and the inference speed can be 40+ times faster.
<div align="center">
<img src="https://user-images.githubusercontent.com/19523330/173011854-b10fcd7a-b799-4dfd-a1cf-9504952a3c44.png" width = "800" />
</div>
The solution mainly includes 4 parts, namely: PP-LCNet lightweight backbone network, SSLD pre-trained model, Ensemble Data Augmentation (EDA) and SKL-UGI knowledge distillation algorithm. In addition, we also adopt the method of hyperparameter search to efficiently optimize the hyperparameters in training. Below, we take the person exists or not scene as an example to illustrate the solution.
**Note**:For some specific scenarios, we provide basic training documents for reference, such as [person exists or not classification model](PULC_person_exists_en.md), etc. You can find these documents [here](./PULC_model_list_en.md). If the methods in these documents do not meet your needs, or if you need a custom training task, you can refer to this document.
<a name="2"></a>
### 2. Data preparation
<a name="2.1"></a>
#### 2.1 Dataset format description
PaddleClas uses the `txt` format file to specify the training set and validation set. Take the person exists or not scene as an example, you need to specify `train_list.txt` and `val_list.txt` as the data labels of the training set and validation set. The format is in the form of as follows:
```
# Each line uses "space" to separate the image path and label
train/1.jpg 0
train/10.jpg 1
...
```
If you want to get more information about common classification datasets, you can refer to the document [PaddleClas Classification Dataset Format Description](../data_preparation/classification_dataset_en.md).
<a name="2.2"></a>
#### 2.2 Annotation file generation method
If you already have the data in the actual scene, you can label it according to the format in the previous section. Here, we provide a script to quickly generate annotation files. You only need to put different categories of data in folders and run the script to generate annotation files.
First, assume that the path where you store the data is `./train`, `train/` contains the data of each category, the category number starts from 0, and the folder of each category contains specific image data.
```shell
train
├── 0
│   ├── 0.jpg
│   ├── 1.jpg
│   └── ...
└── 1
├── 0.jpg
├── 1.jpg
└── ...
└── ...
```
```shell
tree -r -i -f train | grep -E "jpg|JPG|jpeg|JPEG|png|PNG" | awk -F "/" '{print $0" "$2}' > train_list.txt
```
Among them, if more image name suffixes are involved, the content after `grep -E` can be added, and the `2` in `$2` is the level of the category number folder.
**Note:** The above is an introduction to the method of dataset acquisition and generation. Here you can directly download the person exists or not scene data to quickly start the experience.
Go to the PaddleClas directory.
```
cd path_to_PaddleClas
```
Go to the `dataset/` directory, download and unzip the data.
```shell
cd dataset
wget https://paddleclas.bj.bcebos.com/data/PULC/person_exists.tar
tar -xf person_exists.tar
cd ../
```
<a name="3"></a>
### 3. Training with standard classification configuration
<a name="3.1"></a>
#### 3.1 PP-LCNet as backbone
PULC adopts the lightweight backbone network PP-LCNet, which is 50% faster than other networks with the same accuracy. You can view the detailed introduction of the backbone network in [PP-LCNet Introduction](../models/PP-LCNet_en.md).
The command to train with PP-LCNet is:
```shell
export CUDA_VISIBLE_DEVICES=0,1,2,3
python3 -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./ppcls/configs/PULC/person_exists/PPLCNet_x1_0_search.yaml
```
For performance comparison, we also provide configuration files for the large model SwinTransformer_tiny and the lightweight model MobileNetV3_small_x0_35, which you can train with the command:
SwinTransformer_tiny:
```shell
export CUDA_VISIBLE_DEVICES=0,1,2,3
python3 -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./ppcls/configs/PULC/person_exists/SwinTransformer_tiny_patch4_window7_224.yaml
```
MobileNetV3_small_x0_35:
```shell
export CUDA_VISIBLE_DEVICES=0,1,2,3
python3 -m paddle.distributed.launch \
--gpus="0,1,2,3" \
tools/train.py \
-c ./ppcls/configs/PULC/person_exists/MobileNetV3_small_x0_35.yaml
```
The accuracy of the trained models is compared in the following table.
| Model | Tpr(%) | Latency(ms) | Storage Size(M) | Strategy |
|-------|-----------|----------|---------------|---------------|
| SwinTranformer_tiny | 95.69 | 95.30 | 107 | Use ImageNet pretrained model|
| MobileNetV3_small_x0_35 | 68.25 | 2.85 | 1.6 | Use ImageNet pretrained model |
| PPLCNet_x1_0 | 89.57 | 2.12 | 6.5 | Use ImageNet pretrained model |
It can be seen that PP-LCNet is much faster than SwinTransformer, but the accuracy is also slightly lower. Below we improve the accuracy of the PP-LCNet model through a series of optimizations.
<a name="3.2"></a>
#### 3.2 SSLD pretrained model
SSLD is a semi-supervised distillation algorithm developed by Baidu. On the ImageNet dataset, the model accuracy can be improved by 3-7 points. You can find a detailed introduction in [SSLD introduction](../advanced_tutorials/distillation/distillation_en.md). We found that using SSLD pre-trained weights can effectively improve the accuracy of the applied classification model. In addition, using a smaller resolution in training can effectively improve model accuracy. At the same time, we also optimize the learning rate.
Based on the above three improvements, the accuracy of our trained model is 92.1%, an increase of 2.6%.
<a name="3.3"></a>
#### 3.3 EDA strategy
Data augmentation is a commonly used optimization strategy in vision algorithms, which can significantly improve the accuracy of the model. In addition to the traditional RandomCrop, RandomFlip, etc. methods, we also apply RandomAugment and RandomErasing. You can find a detailed introduction at [Data Augmentation Introduction](../advanced_tutorials/DataAugmentation_en.md).
Since these two kinds of data augmentation greatly modify the picture, making the classification task more difficult, it may lead to under-fitting of the model on some datasets. We will set the probability of enabling these two methods in advance.
Based on the above improvements, we obtained a model accuracy of 93.43%, an increase of 1.3%.
<a name="3.4"></a>
#### 3.4 SKL-UGI knowledge distillation
Knowledge distillation is a method that can effectively improve the accuracy of small models. You can find a detailed introduction in [Introduction to Knowledge Distillation](../advanced_tutorials/distillation/distillation_en.md). We choose ResNet101_vd as the teacher model for distillation. In order to adapt to the distillation process, we also adjust the learning rate of different stages of the network here. Based on the above improvements, we trained the model to get a model accuracy of 95.6%, an increase of 1.4%.
<a name="3.5"></a>
#### 3.5 Summary
After the optimization of the above methods, the final accuracy of PP-LCNet reaches 95.6%, reaching the accuracy level of the large model. We summarize the experimental results in the following table:
| Model | Tpr(%) | Latency(ms) | Storage Size(M) | Strategy |
|-------|-----------|----------|---------------|---------------|
| SwinTranformer_tiny | 95.69 | 95.30 | 107 | Use ImageNet pretrained model |
| MobileNetV3_small_x0_35 | 68.25 | 2.85 | 1.6 | Use ImageNet pretrained model |
| PPLCNet_x1_0 | 89.57 | 2.12 | 6.5 | Use ImageNet pretrained model |
| PPLCNet_x1_0 | 92.10 | 2.12 | 6.5 | Use SSLD pretrained model |
| PPLCNet_x1_0 | 93.43 | 2.12 | 6.5 | Use SSLD pretrained model + EDA Strategy|
| <b>PPLCNet_x1_0<b> | <b>95.60<b> | <b>2.12<b> | <b>6.5<b> | Use SSLD pretrained model + EDA Strategy + SKL-UGI knowledge distillation |
We also used the same optimization strategy in the other 8 scenarios and got the following results:
| scenarios | large model | large model metrics(%) | small model | small model metrics(%) |
|----------|----------|----------|----------|----------|
| Pedestrian Attribute Classification | Res2Net200_vd | 81.25 | PPLCNet_x1_0 | 78.59 |
| Classification of Wheather Wearing Safety Helmet | Res2Net200_vd| 98.92 | PPLCNet_x1_0 |99.38 |
| Traffic Sign Classification | SwinTransformer_tiny | 98.11 | PPLCNet_x1_0 | 98.35 |
| Vehicle Attribute Classification | Res2Net200_vd_26w_4s | 91.36 | PPLCNet_x1_0 | 90.81 |
| Car Exists Classification | SwinTransformer_tiny | 97.71 | PPLCNet_x1_0 | 95.92 |
| Text Image Orientation Classification | SwinTransformer_tiny |99.12 | PPLCNet_x1_0 | 99.06 |
| Text-line Orientation Classification | SwinTransformer_tiny | 93.61 | PPLCNet_x1_0 | 96.01 |
| Language Classification | SwinTransformer_tiny | 98.12 | PPLCNet_x1_0 | 99.26 |
It can be seen from the results that the PULC scheme can improve the model accuracy in multiple application scenarios. Using the PULC scheme can greatly reduce the workload of model optimization and quickly obtain models with higher accuracy.
<a name="4"></a>
### 4. Hyperparameter Search
In the above training process, we adjusted parameters such as learning rate, data augmentation probability, and stage learning rate mult list. The optimal values of these parameters may not be the same in different scenarios. We provide a quick hyperparameter search script to automate the process of hyperparameter tuning. This script traverses the parameters in the search value list to replace the parameters in the default configuration, then trains in sequence, and finally selects the parameters corresponding to the model with the highest accuracy as the search result.
<a name="4.1"></a>
#### 4.1 Search based on default configuration
The configuration file [search.yaml](../../../ppcls/configs/PULC/person_exists/search.yaml) defines the configuration of hyperparameter search in person exists or not scenarios. Use the following commands to complete hyperparameter search.
```bash
python3 tools/search_strategy.py -c ppcls/configs/PULC/person_exists/search.yaml
```
**Note**:Regarding the search part, we are also constantly improving, so stay tuned.
<a name="4.2"></a>
#### 4.2 Custom search configuration
You can also modify the configuration of hyperparameter search based on training results or your parameter tuning experience.
Modify the `search_values` field in `lrs` to modify the list of learning rate search values;
Modify the `search_values` field in `resolutions` to modify the search value list of resolutions;
Modify the `search_values` field in `ra_probs` to modify the search value list of RandAugment activation probability;
Modify the `search_values` field in `re_probs` to modify the search value list of RnadomErasing on probability;
Modify the `search_values` field in `lr_mult_list` to modify the lr_mult search value list;
Modify the `search_values` field in `teacher` to modify the search list of the teacher model.
After the search is completed, the final results will be generated in `output/search_person_exists`, where, except for `search_res`, the directories in `output/search_person_exists` are the weights and training log files of the results of the corresponding hyperparameters of each search training, ` search_res` corresponds to the result of knowledge distillation, that is, the final model. The weights of the model are stored in `output/output_dir/search_person_exists/DistillationModel/best_model_student.pdparams`.
# PaddleClas wheel package
Paddleclas supports Python WHL package for prediction. At present, WHL package only supports image classification, but does not support subject detection, feature extraction and vector retrieval.
PaddleClas supports Python wheel package for prediction. At present, PaddleClas wheel supports image classification including ImagetNet1k models and PULC models, but does not support mainbody detection, feature extraction and vector retrieval.
---
......@@ -9,7 +9,7 @@ Paddleclas supports Python WHL package for prediction. At present, WHL package o
- [1. Installation](#1)
- [2. Quick Start](#2)
- [3. Definition of Parameters](#3)
- [4. Usage](#4)
- [4. More usage](#4)
- [4.1 View help information](#4.1)
- [4.2 Prediction using inference model provide by PaddleClas](#4.2)
- [4.3 Prediction using local model files](#4.3)
......@@ -20,6 +20,7 @@ Paddleclas supports Python WHL package for prediction. At present, WHL package o
- [4.8 Specify the mapping between class id and label name](#4.8)
<a name="1"></a>
## 1. Installation
* installing from pypi
......@@ -36,8 +37,14 @@ pip3 install dist/*
```
<a name="2"></a>
## 2. Quick Start
* Using the `ResNet50` model provided by PaddleClas, the following image(`'docs/images/inference_deployment/whl_demo.jpg'`) as an example.
<a name="2.1"></a>
### 2.1 ImageNet1k models
Using the `ResNet50` model provided by PaddleClas, the following image(`'docs/images/inference_deployment/whl_demo.jpg'`) as an example.
![](../../images/inference_deployment/whl_demo.jpg)
......@@ -68,25 +75,89 @@ filename: docs/images/inference_deployment/whl_demo.jpg, top-5, class_ids: [8, 7
Predict complete!
```
<a name="2.2"></a>
### 2.2 PULC models
PULC integrates various state-of-the-art algorithms such as backbone network, data augmentation and distillation, etc., and finally can automatically obtain a lightweight and high-precision image classification model.
PaddleClas provides a series of test cases, which contain demos of different scenes about people, cars, OCR, etc. Click [here](https://paddleclas.bj.bcebos.com/data/PULC/pulc_demo_imgs.zip) to download the data.
Prection using the PULC "Human Exists Classification" model provided by PaddleClas:
* Python
```python
import paddleclas
model = paddleclas.PaddleClas(model_name="person_exists")
result = model.predict(input_data="pulc_demo_imgs/person_exists/objects365_01780782.jpg")
print(next(result))
```
```
>>> result
[{'class_ids': [0], 'scores': [0.9955421453341842], 'label_names': ['nobody'], 'filename': 'pulc_demo_imgs/person_exists/objects365_01780782.jpg'}]
```
`Nobody` means there is no one in the image, `someone` means there is someone in the image. Therefore, the prediction result indicates that there is no one in the figure.
**Note**: `model.predict()` is a generator, so `next()` or `for` is needed to call it. This would to predict by batch that length is `batch_size`, default by 1. You can specify the argument `batch_size` and `model_name` when instantiating PaddleClas object, for example: `model = paddleclas.PaddleClas(model_name="person_exists", batch_size=2)`. Please refer to [Supported Model List](#PULC_Models) for the supported model list.
* CLI
```bash
paddleclas --model_name=person_exists --infer_imgs=pulc_demo_imgs/person_exists/objects365_01780782.jpg
```
```
>>> result
class_ids: [0], scores: [0.9955421453341842], label_names: ['nobody'], filename: pulc_demo_imgs/person_exists/objects365_01780782.jpg
Predict complete!
```
**Note**: The "--infer_imgs" argument specify the image(s) to be predict, and you can also specify a directoy contains images. If use other model, you can specify the `--model_name` argument. Please refer to [Supported Model List](#PULC_Models) for the supported model list.
<a name="PULC_Models"></a>
**Supported Model List**
The name of PULC series models are as follows:
| Name | Intro |
| --- | --- |
| person_exists | Human Exists Classification |
| person_attribute | Pedestrian Attribute Classification |
| safety_helmet | Classification of Wheather Wearing Safety Helmet |
| traffic_sign | Traffic Sign Classification |
| vehicle_attribute | Vehicle Attribute Classification |
| car_exists | Car Exists Classification |
| text_image_orientation | Text Image Orientation Classification |
| textline_orientation | Text-line Orientation Classification |
| language_classification | Language Classification |
Please refer to [Human Exists Classification](../PULC/PULC_person_exists_en.md)[Pedestrian Attribute Classification](../PULC/PULC_person_attribute_en.md)[Classification of Wheather Wearing Safety Helmet](../PULC/PULC_safety_helmet_en.md)[Traffic Sign Classification](../PULC/PULC_traffic_sign_en.md)[Vehicle Attribute Classification](../PULC/PULC_vehicle_attribute_en.md)[Car Exists Classification](../PULC/PULC_car_exists_en.md)[Text Image Orientation Classification](../PULC/PULC_text_image_orientation_en.md)[Text-line Orientation Classification](../PULC/PULC_textline_orientation_en.md)[Language Classification](../PULC/PULC_language_classification_en.md) for more information about different scenarios.
<a name="3"></a>
## 3. Definition of Parameters
The following parameters can be specified in Command Line or used as parameters of the constructor when instantiating the PaddleClas object in Python.
* model_name(str): If using inference model based on ImageNet1k provided by Paddle, please specify the model's name by the parameter.
* inference_model_dir(str): Local model files directory, which is valid when `model_name` is not specified. The directory should contain `inference.pdmodel` and `inference.pdiparams`.
* infer_imgs(str): The path of image to be predicted, or the directory containing the image files, or the URL of the image from Internet.
* use_gpu(bool): Whether to use GPU or not, default by `True`.
* gpu_mem(int): GPU memory usages,default by `8000`
* use_tensorrt(bool): Whether to open TensorRT or not. Using it can greatly promote predict preformance, default by `False`.
* enable_mkldnn(bool): Whether enable MKLDNN or not, default `False`.
* cpu_num_threads(int): Assign number of cpu threads, valid when `--use_gpu` is `False` and `--enable_mkldnn` is `True`, default by `10`.
* batch_size(int): Batch size, default by `1`.
* resize_short(int): Resize the minima between height and width into `resize_short`, default by `256`.
* crop_size(int): Center crop image to `crop_size`, default by `224`.
* topk(int): Print (return) the `topk` prediction results, default by `5`.
* class_id_map_file(str): The mapping file between class ID and label, default by `ImageNet1K` dataset's mapping.
* pre_label_image(bool): whether prelabel or not, default=False.
* save_dir(str): The directory to save the prediction results that can be used as pre-label, default by `None`, that is, not to save.
* use_gpu(bool): Whether to use GPU or not.
* gpu_mem(int): GPU memory usages.
* use_tensorrt(bool): Whether to open TensorRT or not. Using it can greatly promote predict preformance.
* enable_mkldnn(bool): Whether enable MKLDNN or not.
* cpu_num_threads(int): Assign number of cpu threads, valid when `--use_gpu` is `False` and `--enable_mkldnn` is `True`.
* batch_size(int): Batch size.
* resize_short(int): Resize the minima between height and width into `resize_short`.
* crop_size(int): Center crop image to `crop_size`.
* topk(int): Print (return) the `topk` prediction results when Topk postprocess is used.
* threshold(float): The threshold of ThreshOutput when postprocess is used.
* class_id_map_file(str): The mapping file between class ID and label.
* save_dir(str): The directory to save the prediction results that can be used as pre-label.
**Note**: If you want to use `Transformer series models`, such as `DeiT_***_384`, `ViT_***_384`, etc., please pay attention to the input size of model, and need to set `resize_short=384`, `resize=384`. The following is a demo.
......@@ -103,6 +174,7 @@ clas = PaddleClas(model_name='ViT_base_patch16_384', resize_short=384, crop_size
```
<a name="4"></a>
## 4. Usage
PaddleClas provides two ways to use:
......@@ -110,6 +182,7 @@ PaddleClas provides two ways to use:
2. Bash command line programming.
<a name="4.1"></a>
### 4.1 View help information
* CLI
......@@ -118,6 +191,7 @@ paddleclas -h
```
<a name="4.2"></a>
### 4.2 Prediction using inference model provide by PaddleClas
You can use the inference model provided by PaddleClas to predict, and only need to specify `model_name`. In this case, PaddleClas will automatically download files of specified model and save them in the directory `~/.paddleclas/`.
......@@ -136,6 +210,7 @@ paddleclas --model_name='ResNet50' --infer_imgs='docs/images/inference_deploymen
```
<a name="4.3"></a>
### 4.3 Prediction using local model files
You can use the local model files trained by yourself to predict, and only need to specify `inference_model_dir`. Note that the directory must contain `inference.pdmodel` and `inference.pdiparams`.
......@@ -154,6 +229,7 @@ paddleclas --inference_model_dir='./inference/' --infer_imgs='docs/images/infere
```
<a name="4.4"></a>
### 4.4 Prediction by batch
You can predict by batch, only need to specify `batch_size` when `infer_imgs` is direcotry contain image files.
......@@ -173,6 +249,7 @@ paddleclas --model_name='ResNet50' --infer_imgs='docs/images/' --batch_size 2
```
<a name="4.5"></a>
### 4.5 Prediction of Internet image
You can predict the Internet image, only need to specify URL of Internet image by `infer_imgs`. In this case, the image file will be downloaded and saved in the directory `~/.paddleclas/images/`.
......@@ -191,6 +268,7 @@ paddleclas --model_name='ResNet50' --infer_imgs='https://raw.githubusercontent.c
```
<a name="4.6"></a>
### 4.6 Prediction of NumPy.array format image
In Python code, you can predict the `NumPy.array` format image, only need to use the `infer_imgs` to transfer variable of image data. Note that the models in PaddleClas only support to predict 3 channels image data, and channels order is `RGB`.
......@@ -205,6 +283,7 @@ print(next(result))
```
<a name="4.7"></a>
### 4.7 Save the prediction result(s)
You can save the prediction result(s) as pre-label, only need to use `pre_label_out_dir` to specify the directory to save.
......@@ -223,6 +302,7 @@ paddleclas --model_name='ResNet50' --infer_imgs='docs/images/' --save_dir='./out
```
<a name="4.8"></a>
### 4.8 Specify the mapping between class id and label name
You can specify the mapping between class id and label name, only need to use `class_id_map_file` to specify the mapping file. PaddleClas uses ImageNet1K's mapping by default.
......
# PULC 有人/无人分类模型
# PULC 有车/无车分类模型
------
......@@ -40,7 +40,7 @@
## 1. 模型和应用场景介绍
该案例提供了用户使用 PaddleClas 的超轻量图像分类方案(PULC,Practical Ultra Lightweight Classification)快速构建轻量级、高精度、可落地的有人/无人的分类模型。该模型可以广泛应用于如监控场景、人员进出管控场景、海量数据过滤场景等。
该案例提供了用户使用 PaddleClas 的超轻量图像分类方案(PULC,Practical Ultra Lightweight Classification)快速构建轻量级、高精度、可落地的有车/无车的分类模型。该模型可以广泛应用于如监控场景、海量数据过滤场景等。
下表列出了判断图片中是否有车的二分类模型的相关指标,前两行展现了使用 SwinTranformer_tiny 和 MobileNetV3_small_x0_35 作为 backbone 训练得到的模型的相关指标,第三行至第六行依次展现了替换 backbone 为 PPLCNet_x1_0、使用 SSLD 预训练模型、使用 SSLD 预训练模型 + EDA 策略、使用 SSLD 预训练模型 + EDA 策略 + SKL-UGI 知识蒸馏策略训练得到的模型的相关指标。
......@@ -58,7 +58,7 @@
**备注:**
* `Tpr`指标的介绍可以参考 [3.2 小节](#3.2)的备注部分,延时是基于 Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz 测试得到,开启 MKLDNN 加速策略,线程数为10。
* `Tpr`指标的介绍可以参考 [3.3节](#3.3)的备注部分,延时是基于 Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz 测试得到,开启 MKLDNN 加速策略,线程数为10。
* 关于PP-LCNet的介绍可以参考[PP-LCNet介绍](../models/PP-LCNet.md),相关论文可以查阅[PP-LCNet paper](https://arxiv.org/abs/2109.15099)
......@@ -160,7 +160,7 @@ print(next(result))
- 训练集合,本案例处理了 Objects365 数据训练集的标注文件,如果某张图含有“car”的标签,且这个框的面积在整张图中的比例大于 10%,即认为该张图中含有车,如果某张图中没有任何与交通工具,例如car、bus等相关的的标签,则认为该张图中不含有车。经过处理后,得到 108629 条可用数据,其中有车的数据有 27422 条,无车的数据 81207 条。
- 验证集合,处理方法与训练集相同,数据来源 Objects365 数据集的验证集。为了测试结果准确,验证集经过人工校正,去除了一些可能存在标注错误的图像。
- 验证集合,处理方法与训练集相同,数据来源 Objects365 数据集的验证集。为了测试结果准确,验证集经过人工校正,去除了一些可能存在标注错误的图像。
* 注:由于objects365的标签并不是完全互斥的,例如F1赛车可能是 "F1 Formula",也可能被标称"car"。为了减轻干扰,我们仅保留"car"标签作为有车,而将不含任何交通工具的图作为无车。
......@@ -265,7 +265,7 @@ python3 tools/infer.py \
输出结果如下:
```
[{'class_ids': [1], 'scores': [0.9871138], 'label_names': ['contains_vehicle'], 'filename': 'deploy/images/PULC/car_exists/objects365_00001507.jpeg'}]
[{'class_ids': [1], 'scores': [0.9871138], 'label_names': ['contains_car'], 'filename': 'deploy/images/PULC/car_exists/objects365_00001507.jpeg'}]
```
**备注:**
......@@ -274,7 +274,7 @@ python3 tools/infer.py \
* 默认是对 `deploy/images/PULC/car_exists/objects365_00001507.jpeg` 进行预测,此处也可以通过增加字段 `-o Infer.infer_imgs=xxx` 对其他图片预测。
* 二分类默认的阈值为0.5, 如果需要指定阈值,可以重写 `Infer.PostProcess.threshold` ,如`-o Infer.PostProcess.threshold=0.9794`,该值需要根据实际场景来确定,此处的 `0.9794` 是在该场景中的 `val` 数据集在分之一 Fpr 下得到的最佳 Tpr 所得到的。
* 二分类默认的阈值为0.5, 如果需要指定阈值,可以重写 `Infer.PostProcess.threshold` ,如`-o Infer.PostProcess.threshold=0.9794`,该值需要根据实际场景来确定,此处的 `0.9794` 是在该场景中的 `val` 数据集在分之一 Fpr 下得到的最佳 Tpr 所得到的。
<a name="4"></a>
......@@ -326,7 +326,7 @@ python3 -m paddle.distributed.launch \
## 5. 超参搜索
[3.2 节](#3.2)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
[3.3 节](#3.3)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
**备注:** 此部分内容是可选内容,搜索过程需要较长的时间,您可以根据自己的硬件情况来选择执行。如果没有更换数据集,可以忽略此节内容。
......@@ -417,7 +417,7 @@ objects365_00001507.jpeg: class id(s): [1], score(s): [0.99], label_name(s
```
**备注:** 二分类默认的阈值为0.5, 如果需要指定阈值,可以重写 `Infer.PostProcess.threshold` ,如`-o Infer.PostProcess.threshold=0.9794`,该值需要根据实际场景来确定,此处的 `0.9794` 是在该场景中的 `val` 数据集在分之一 Fpr 下得到的最佳 Tpr 所得到的。该阈值的确定方法可以参考[3.3节](#3.3)备注部分。
**备注:** 二分类默认的阈值为0.5, 如果需要指定阈值,可以重写 `Infer.PostProcess.threshold` ,如`-o Infer.PostProcess.threshold=0.9794`,该值需要根据实际场景来确定,此处的 `0.9794` 是在该场景中的 `val` 数据集在分之一 Fpr 下得到的最佳 Tpr 所得到的。该阈值的确定方法可以参考[3.3节](#3.3)备注部分。
<a name="6.2.2"></a>
......
......@@ -328,7 +328,7 @@ python3 -m paddle.distributed.launch \
## 5. 超参搜索
[3.2 节](#3.2)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
[3.3 节](#3.3)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
**备注:** 此部分内容是可选内容,搜索过程需要较长的时间,您可以根据自己的硬件情况来选择执行。如果没有更换数据集,可以忽略此节内容。
......
......@@ -2,7 +2,7 @@
------
本文主要介绍PaddleClas whl包对 PULC 系列模型的快速使用
本文主要介绍通过 PaddleClas whl 包,使用 PULC 系列模型进行预测
## 目录
......@@ -49,9 +49,7 @@ pip3 install paddleclas
## 2. 快速体验
PaddleClas 提供了一系列测试图片,里边包含人、车、OCR等方向的多个场景大的demo数据。点击[这里](https://paddleclas.bj.bcebos.com/data/PULC/pulc_demo_imgs.zip)下载并解压,然后在终端中切换到相应目录。
PaddleClas 提供了一系列测试图片,里边包含人、车、OCR等方向的多个场景的demo数据。点击[这里](https://paddleclas.bj.bcebos.com/data/PULC/pulc_demo_imgs.zip)下载并解压,然后在终端中切换到相应目录。
<a name="2.1"></a>
......@@ -125,7 +123,3 @@ PULC 系列模型的名称和简介如下:
通过本节内容,相信您已经熟练掌握 PaddleClas whl 包的 PULC 模型使用方法并获得了初步效果。
PULC 方法产出的系列模型在人、车、OCR等方向的多个场景中均验证有效,用超轻量模型就可实现与 SwinTransformer 模型接近的精度,预测速度提高 40+ 倍。并且打通数据、模型训练、压缩和推理部署全流程,具体地,您可以参考[PULC有人/无人分类模型](PULC_person_exists.md)[PULC人体属性识别模型](PULC_person_attribute.md)[PULC佩戴安全帽分类模型](PULC_safety_helmet.md)[PULC交通标志分类模型](PULC_traffic_sign.md)[PULC车辆属性识别模型](PULC_vehicle_attribute.md)[PULC有车/无车分类模型](PULC_car_exists.md)[PULC含文字图像方向分类模型](PULC_text_image_orientation.md)[PULC文本行方向分类模型](PULC_textline_orientation.md)[PULC语种分类模型](PULC_language_classification.md)
......@@ -165,7 +165,7 @@ print(next(result))
处理后的数据集部分数据可视化如下:
![](../../images/PULC/docs/safety_helmet_data_demo.png)
![](../../images/PULC/docs/safety_helmet_data_demo.jpg)
此处提供了经过上述方法处理好的数据,可以直接下载得到。
......@@ -217,7 +217,7 @@ python3 -m paddle.distributed.launch \
-c ./ppcls/configs/PULC/safety_helmet/PPLCNet_x1_0.yaml
```
验证集的最佳指标在 `0.975-0.985` 之间(数据集较小,容易造成波动)。
验证集的最佳指标在 `0.985-0.993` 之间(数据集较小,容易造成波动)。
**备注:**
......
......@@ -55,11 +55,11 @@
| <b>PPLCNet_x1_0**<b> | <b>96.01<b> | <b>2.72<b> | <b>6.5<b> | 使用 SSLD 预训练模型+EDA 策略|
| PPLCNet_x1_0** | 95.86 | 2.72 | 6.5 | 使用 SSLD 预训练模型+EDA 策略+SKL-UGI 知识蒸馏策略|
从表中可以看出,backbone 为 SwinTranformer_tiny 时精度较高,但是推理速度较慢。将 backboone 替换为轻量级模型 MobileNetV3_small_x0_35 后,速度可以大幅提升,精度下降也比较明显。将 backbone 替换为 PPLCNet_x1_0 时,精度较 MobileNetV3_small_x0_35 高 8.6 个百分点,速度快10%左右。在此基础上,更改分辨率和stride, 速度变慢 27%,但是精度可以提升 4.5%(采用[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)的方案),使用 SSLD 预训练模型后,精度可以继续提升约 0.05% ,进一步地,当融合EDA策略后,精度可以再提升 1.9 个百分点。最后,融合SKL-UGI 知识蒸馏策略后,在该场景无效。关于 PULC 的训练方法和推理部署方法将在下面详细介绍。
从表中可以看出,backbone 为 SwinTranformer_tiny 时精度较高,但是推理速度较慢。将 backboone 替换为轻量级模型 MobileNetV3_small_x0_35 后,速度可以大幅提升,精度下降也比较明显。将 backbone 替换为 PPLCNet_x1_0 时,精度较 MobileNetV3_small_x0_35 高 8.6 个百分点,速度快10%左右。在此基础上,更改分辨率和stride, 速度变慢 27%,但是精度可以提升 4.5 个百分点(采用[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)的方案),使用 SSLD 预训练模型后,精度可以继续提升约 0.05 个百分点 ,进一步地,当融合EDA策略后,精度可以再提升 1.9 个百分点。最后,融合SKL-UGI 知识蒸馏策略后,在该场景无效。关于 PULC 的训练方法和推理部署方法将在下面详细介绍。
**备注:**
* 其中不带\*的模型表示分辨率为224x224,带\*的模型表示分辨率为48x192(h*w),数据增强从网络中的 stride 改为 `[2, [2, 1], [2, 1], [2, 1], [2, 1]]`,其中,外层列表中的每一个元素代表网络结构下采样层的stride,该策略为 [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) 提供的文本行方向分类器方案。带\*\*的模型表示分辨率为80x160(h*w), 网络中的 stride 改为 `[2, [2, 1], [2, 1], [2, 1], [2, 1]]`,其中,外层列表中的每一个元素代表网络结构下采样层的stride,此分辨率是经过[SHAS 超参数搜索策略](#TODO)搜索得到的。
* 其中不带\*的模型表示分辨率为224x224,带\*的模型表示分辨率为48x192(h\*w),数据增强从网络中的 stride 改为 `[2, [2, 1], [2, 1], [2, 1], [2, 1]]`,其中,外层列表中的每一个元素代表网络结构下采样层的stride,该策略为 [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) 提供的文本行方向分类器方案。带\*\*的模型表示分辨率为80x160(h\*w), 网络中的 stride 改为 `[2, [2, 1], [2, 1], [2, 1], [2, 1]]`,其中,外层列表中的每一个元素代表网络结构下采样层的stride,此分辨率是经过[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)搜索得到的。
* 延时是基于 Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz 测试得到,开启 MKLDNN 加速策略,线程数为10。
* 关于PP-LCNet的介绍可以参考[PP-LCNet介绍](../models/PP-LCNet.md),相关论文可以查阅[PP-LCNet paper](https://arxiv.org/abs/2109.15099)
......@@ -116,7 +116,6 @@ Predict complete!
**备注**: 更换其他预测的数据时,只需要改变 `--infer_imgs=xx` 中的字段即可,支持传入整个文件夹。
* 在 Python 代码中预测
```python
import paddleclas
......@@ -140,7 +139,7 @@ print(next(result))
### 3.1 环境配置
* 安装:请先参考 [Paddle 安装教程](../installation/install_paddle.md) 以及 [PaddleClas 安装教程](../installation/install_paddleclas.md) 配置 PaddleClas 运行环境。
* 安装:请先参考文档 [环境准备](../installation/install_paddleclas.md) 配置 PaddleClas 运行环境。
<a name="3.2"></a>
......@@ -168,17 +167,15 @@ print(next(result))
![](../../images/PULC/docs/textline_orientation_data_demo.png)
此处提供了经过上述方法处理好的数据,可以直接下载得到。
进入 PaddleClas 目录。
```
cd path_to_PaddleClas
```
进入 `dataset/` 目录,下载并解压有人/无人场景的数据。
进入 `dataset/` 目录,下载并解压文本行方向分类场景的数据。
```shell
cd dataset
......@@ -190,7 +187,6 @@ cd ../
执行上述命令后,`dataset/` 下存在 `textline_orientation` 目录,该目录中具有以下数据:
```
├── 0
│   ├── img_0.jpg
│   ├── img_1.jpg
......@@ -253,7 +249,7 @@ python3 tools/eval.py \
```python
python3 tools/infer.py \
-c ./ppcls/configs/PULC/textline_orientation/PPLCNet_x1_0.yaml \
-o Global.pretrained_model=output/PPLCNet_x1_0/best_model \
-o Global.pretrained_model=output/PPLCNet_x1_0/best_model
```
输出结果如下:
......@@ -318,7 +314,7 @@ python3 -m paddle.distributed.launch \
## 5. 超参搜索
[3.2 节](#3.2)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
[3.3 节](#3.3)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
**备注:** 此部分内容是可选内容,搜索过程需要较长的时间,您可以根据自己的硬件情况来选择执行。
......
......@@ -212,7 +212,7 @@ SSLD 是百度自研的半监督蒸馏算法,在 ImageNet 数据集上,模
#### 4.1 基于默认配置搜索
配置文件 [search.yaml](../../ppcls/configs/PULC/person_exists/search.yaml) 定义了有人/无人场景超参搜索的配置,使用如下命令即可完成超参数的搜索。
配置文件 [search.yaml](../../../ppcls/configs/PULC/person_exists/search.yaml) 定义了有人/无人场景超参搜索的配置,使用如下命令即可完成超参数的搜索。
```bash
python3 tools/search_strategy.py -c ppcls/configs/PULC/person_exists/search.yaml
......
......@@ -55,7 +55,7 @@
| PPLCNet_x1_0 | 90.59 | 2.36 | 7.2 | 使用SSLD预训练模型+EDA策略|
| <b>PPLCNet_x1_0<b> | <b>90.81<b> | <b>2.36<b> | <b>8.2<b> | 使用SSLD预训练模型+EDA策略+SKL-UGI知识蒸馏策略|
从表中可以看出,backbone 为 Res2Net200_vd_26w_4s 时精度较高,但是推理速度较慢。将 backbone 替换为轻量级模型 MobileNetV3_small_x0_35 后,速度可以大幅提升,但是精度下降明显。将 backbone 替换为 PPLCNet_x1_0 时,精度提升 2.16%,同时速度也提升 23% 左右。在此基础上,使用 SSLD 预训练模型后,在不改变推理速度的前提下,精度可以提升约 0.5%,进一步地,当融合EDA策略后,精度可以再提升 0.52%,最后,在使用 SKL-UGI 知识蒸馏后,精度可以继续提升 0.23%。此时,PPLCNet_x1_0 的精度与 Res2Net200_vd_26w_4s 仅相差0.55%,但是速度快32倍。关于 PULC 的训练方法和推理部署方法将在下面详细介绍。
从表中可以看出,backbone 为 Res2Net200_vd_26w_4s 时精度较高,但是推理速度较慢。将 backbone 替换为轻量级模型 MobileNetV3_small_x0_35 后,速度可以大幅提升,但是精度下降明显。将 backbone 替换为 PPLCNet_x1_0 时,精度提升 2 个百分点,同时速度也提升 23% 左右。在此基础上,使用 SSLD 预训练模型后,在不改变推理速度的前提下,精度可以提升约 0.5 个百分点,进一步地,当融合EDA策略后,精度可以再提升 0.52 个百分点,最后,在使用 SKL-UGI 知识蒸馏后,精度可以继续提升 0.23 个百分点。此时,PPLCNet_x1_0 的精度与 Res2Net200_vd_26w_4s 仅相差 0.55 个百分点,但是速度快 32 倍。关于 PULC 的训练方法和推理部署方法将在下面详细介绍。
**备注:**
......@@ -337,7 +337,7 @@ python3 -m paddle.distributed.launch \
## 5. 超参搜索
[3.2 节](#3.2)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
[3.3 节](#3.3)[4.1 节](#4.1)所使用的超参数是根据 PaddleClas 提供的 `SHAS 超参数搜索策略` 搜索得到的,如果希望在自己的数据集上得到更好的结果,可以参考[SHAS 超参数搜索策略](PULC_train.md#4-超参搜索)来获得更好的训练超参数。
**备注:** 此部分内容是可选内容,搜索过程需要较长的时间,您可以根据自己的硬件情况来选择执行。如果没有更换数据集,可以忽略此节内容。
......@@ -368,10 +368,10 @@ python3 tools/export_model.py \
执行完该脚本后会在 `deploy/models/` 下生成 `PPLCNet_x1_0_vehicle_attributeibute_infer` 文件夹,`models` 文件夹下应有如下文件结构:
```
── PPLCNet_x1_0_vehicle_attribute_infer
├── inference.pdiparams
├── inference.pdiparams.info
└── inference.pdmodel
── PPLCNet_x1_0_vehicle_attribute_infer
├── inference.pdiparams
├── inference.pdiparams.info
└── inference.pdmodel
```
**备注:** 此处的最佳权重是经过知识蒸馏后的权重路径,如果没有执行知识蒸馏的步骤,最佳模型保存在`output/PPLCNet_x1_0/best_model.pdparams`中。
......
......@@ -6,7 +6,7 @@
- [1. 算法介绍](#1)
- [1.1 知识蒸馏简介](#1.1)
- [1.2 SSLD蒸馏策略](#1.2)
- [1.2 SKL-UGI蒸馏策略](#1.3)
- [1.3 SKL-UGI蒸馏策略](#1.3)
- [2. SSLD预训练模型库](#2)
- [3. SSLD使用](#3)
- [3.1 加载SSLD模型进行微调](#3.1)
......@@ -19,6 +19,8 @@
## 1. 算法介绍
<a name="1.1"></a>
### 1.1 简介
PaddleClas 融合已有的知识蒸馏方法 [2,3],提供了一种简单的半监督标签知识蒸馏方案(SSLD,Simple Semi-supervised Label Distillation),基于 ImageNet1k 分类数据集,在 ResNet_vd 以及 MobileNet 系列上的精度均有超过 3% 的绝对精度提升,具体指标如下图所示。
......@@ -27,6 +29,8 @@ PaddleClas 融合已有的知识蒸馏方法 [2,3],提供了一种简单的半
<img src="../../images/distillation/distillation_perform_s.jpg" width = "800" />
</div>
<a name="1.2"></a>
### 1.2 SSLD蒸馏策略
SSLD 的流程图如下图所示。
......@@ -72,6 +76,8 @@ SSLD 蒸馏方案的一大特色就是无需使用图像的真值标签,因此
<a name="1.3"></a>
## 1.3 SKL-UGI蒸馏策略
此外,在无标注数据选择的过程中,我们发现使用更加通用的数据,即使不需要严格的数据筛选过程,也可以帮助知识蒸馏任务获得稳定的精度提升,因而提出了SKL-UGI (Symmetrical-KL Unlabeled General Images distillation)知识蒸馏方案。
通用数据可以使用ImageNet数据或者与场景相似的数据集。更多关于SKL-UGI的应用,请参考:[超轻量图像分类方案PULC使用教程](../PULC/PULC_train.md)
......
......@@ -175,11 +175,11 @@ IMN_MODEL_SERIES = {
]
}
PULC_MODEL_BASE_DOWNLOAD_URL = "https://paddleclas.bj.bcebos.com/models/PULC/{}_infer.tar"
PULC_MODEL_BASE_DOWNLOAD_URL = "https://paddleclas.bj.bcebos.com/models/PULC/inference/{}_infer.tar"
PULC_MODELS = [
"person_exists", "person_attribute", "safety_helmet", "traffic_sign",
"vehicle_exists", "vehicle_attribute", "textline_orientation",
"text_image_orientation", "language_classification"
"car_exists", "language_classification", "person_attribute",
"person_exists", "safety_helmet", "text_image_orientation",
"textline_orientation", "traffic_sign", "vehicle_attribute"
]
......@@ -298,7 +298,7 @@ def args_cfg():
parser.add_argument(
"--use_tensorrt",
type=str2bool,
help="Whether use TensorRT to accelerate. ")
help="Whether use TensorRT to accelerate.")
parser.add_argument(
"--use_fp16", type=str2bool, help="Whether use FP16 to predict.")
parser.add_argument("--batch_size", type=int, help="Batch size.")
......
......@@ -233,7 +233,7 @@ class ShuffleNet(Layer):
elif scale == 1.5:
stage_out_channels = [-1, 24, 176, 352, 704, 1024]
elif scale == 2.0:
stage_out_channels = [-1, 24, 224, 488, 976, 2048]
stage_out_channels = [-1, 24, 244, 488, 976, 2048]
else:
raise NotImplementedError("This scale size:[" + str(scale) +
"] is not implemented!")
......
......@@ -125,8 +125,8 @@ Infer:
PostProcess:
name: ThreshOutput
threshold: 0.5
label_0: nobody
label_1: someone
label_0: no_car
label_1: contains_car
Metric:
Train:
......
......@@ -138,8 +138,8 @@ Infer:
PostProcess:
name: ThreshOutput
threshold: 0.9
label_0: nobody
label_1: someone
label_0: no_car
label_1: contains_car
Metric:
Train:
......
......@@ -154,8 +154,8 @@ Infer:
PostProcess:
name: ThreshOutput
threshold: 0.5
label_0: nobody
label_1: someone
label_0: no_car
label_1: contains_car
Metric:
Train:
......
......@@ -138,8 +138,8 @@ Infer:
PostProcess:
name: ThreshOutput
threshold: 0.5
label_0: nobody
label_1: someone
label_0: no_car
label_1: contains_car
Metric:
Train:
......
......@@ -155,8 +155,8 @@ Infer:
PostProcess:
name: ThreshOutput
threshold: 0.5
label_0: nobody
label_1: someone
label_0: no_car
label_1: contains_car
Metric:
Train:
......
......@@ -36,7 +36,7 @@ Optimizer:
momentum: 0.9
lr:
name: Cosine
learning_rate: 0.4
learning_rate: 0.08
warmup_epoch: 5
regularizer:
name: 'L2'
......@@ -66,11 +66,11 @@ DataLoader:
sampler:
name: DistributedBatchSampler
batch_size: 64
batch_size: 512
drop_last: False
shuffle: True
loader:
num_workers: 4
num_workers: 8
use_shared_memory: True
Eval:
......
......@@ -54,7 +54,7 @@ DataLoader:
dataset:
name: ImageNetDataset
image_root: ./dataset/
cls_label_path: ./dataset/tt100k_clas_v2/label_list_train.txt
cls_label_path: ./dataset/traffic_sign/label_list_train.txt
delimiter: "\t"
transform_ops:
- DecodeImage:
......
......@@ -118,8 +118,7 @@ DataLoader:
use_shared_memory: True
Infer:
# infer_imgs: dataset/traffic_sign_demo/
infer_imgs: dataset/tt100k_clas_v2/test/
infer_imgs: deploy/images/PULC/traffic_sign/99603_17806.jpg
batch_size: 10
transforms:
- DecodeImage:
......
......@@ -113,7 +113,7 @@ DataLoader:
dataset:
name: ImageNetDataset
image_root: ./dataset/
cls_label_path: ./dataset/tt100k_clas_v2/label_list_test.txt
cls_label_path: ./dataset/traffic_sign/label_list_test.txt
delimiter: "\t"
transform_ops:
- DecodeImage:
......
......@@ -123,7 +123,7 @@ DataLoader:
use_shared_memory: True
Infer:
infer_imgs: ./deploy/images/PULC/vehicle_attr/0002_c002_00030670_0.jpg
infer_imgs: ./deploy/images/PULC/vehicle_attribute/0002_c002_00030670_0.jpg
batch_size: 10
transforms:
- DecodeImage:
......
......@@ -38,13 +38,16 @@ setup(
version='0.0.0',
install_requires=requirements,
license='Apache License 2.0',
description='Awesome Image Classification toolkits based on PaddlePaddle ',
description='A treasure chest for visual recognition powered by PaddlePaddle.',
long_description=readme(),
long_description_content_type='text/markdown',
url='https://github.com/PaddlePaddle/PaddleClas',
download_url='https://github.com/PaddlePaddle/PaddleClas.git',
keywords=[
'A treasure chest for image classification powered by PaddlePaddle.'
'image-classification', 'image-recognition', 'pretrained-models',
'knowledge-distillation', 'product-recognition', 'autoaugment',
'cutmix', 'randaugment', 'gridmask', 'deit', 'repvgg',
'swin-transformer', 'image-retrieval-system'
],
classifiers=[
'Intended Audience :: Developers',
......
Markdown is supported
0% .
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