Skip to content

P
PaddleDetection

PaddlePaddle / PaddleDetection
大约 1 年 前同步成功

通知 695
Star 11112
Fork 2696
P
PaddleDetection
未验证 提交 38f4a27f 编写于 作者: W wangxinxin08 提交者： GitHub
浏览文件
操作

[cherry-pick]fix some bugs and polish docs (#6757) 

* fix some bugs and polish docs

* polish docs

* modify develop to release/2.5

* modify docs, test=document_fix

* modify docs, test=document_fix
上级 94861c24
隐藏空白更改
内联 并排
Showing with 136 addition and 88 deletion
configs/rotate/README.md
浏览文件 @ 38f4a27f
...@@ -15,11 +15,12 @@ ...@@ -15,11 +15,12 @@
| 模型 | mAP | 学习率策略 | 角度表示 | 数据增广 | GPU数目 | 每GPU图片数目 | 模型下载 | 配置文件 | | 模型 | mAP | 学习率策略 | 角度表示 | 数据增广 | GPU数目 | 每GPU图片数目 | 模型下载 | 配置文件 |
|:---:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:| |:---:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:|
| [S2ANet](./s2anet/README.md) | 74.0 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/dota/s2anet_alignconv_2x_dota.yml) | | [S2ANet](./s2anet/README.md) | 73.84 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**注意:** **注意:**
- 如果**GPU卡数**或者**batch size**发生了改变，你需要按照公式 **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)** 调整学习率。 - 如果**GPU卡数**或者**batch size**发生了改变，你需要按照公式 **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)** 调整学习率。
- 模型库中的模型默认使用单尺度训练单尺度测试。如果数据增广一栏标明MS，意味着使用多尺度训练和多尺度测试。如果数据增广一栏标明RR，意味着使用RandomRotate数据增广进行训练。
## 数据准备 ## 数据准备
### DOTA数据准备 ### DOTA数据准备
...@@ -36,8 +37,15 @@ ${DOTA_ROOT} ...@@ -36,8 +37,15 @@ ${DOTA_ROOT}
└── labelTxt └── labelTxt
``` ```
DOTA数据集分辨率较高，因此一般在训练和测试之前对图像进行切图，使用单尺度进行切图可以使用以下命令： 对于有标注的数据，每一张图片会对应一个同名的txt文件，文件中每一行为一个旋转框的标注，其格式如下：
``` ```
x1 y1 x2 y2 x3 y3 x4 y4 class_name difficult
```
### 单尺度切图
DOTA数据集分辨率较高，因此一般在训练和测试之前对图像进行离线切图，使用单尺度进行切图可以使用以下命令：
``` bash
# 对于有标注的数据进行切图
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \ --input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \
--output_dir ${OUTPUT_DIR}/trainval1024/ \ --output_dir ${OUTPUT_DIR}/trainval1024/ \
...@@ -45,26 +53,39 @@ python configs/rotate/tools/prepare_data.py \ ...@@ -45,26 +53,39 @@ python configs/rotate/tools/prepare_data.py \
--subsize 1024 \ --subsize 1024 \
--gap 200 \ --gap 200 \
--rates 1.0 --rates 1.0
# 对于无标注的数据进行切图需要设置--image_only
python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/test/ \
--output_dir ${OUTPUT_DIR}/test1024/ \
--coco_json_file DOTA_test1024.json \
--subsize 1024 \
--gap 200 \
--rates 1.0 \
--image_only
``` ```
### 多尺度切图
使用多尺度进行切图可以使用以下命令： 使用多尺度进行切图可以使用以下命令：
``` ``` bash
# 对于有标注的数据进行切图
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \ --input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \
--output_dir ${OUTPUT_DIR}/trainval/ \ --output_dir ${OUTPUT_DIR}/trainval/ \
--coco_json_file DOTA_trainval1024.json \ --coco_json_file DOTA_trainval1024.json \
--subsize 1024 \ --subsize 1024 \
--gap 500 \ --gap 500 \
--rates 0.5 1.0 1.5 \ --rates 0.5 1.0 1.5
```
对于无标注的数据可以设置`--image_only`进行切图，如下所示： # 对于无标注的数据进行切图需要设置--image_only
```
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/test/ \ --input_dirs ${DOTA_ROOT}/test/ \
--output_dir ${OUTPUT_DIR}/test1024/ \ --output_dir ${OUTPUT_DIR}/test1024/ \
--coco_json_file DOTA_test1024.json \ --coco_json_file DOTA_test1024.json \
--subsize 1024 \ --subsize 1024 \
--gap 200 \ --gap 500 \
--rates 1.0 \ --rates 0.5 1.0 1.5 \
--image_only --image_only
``` ```
......
configs/rotate/README_en.md
浏览文件 @ 38f4a27f
...@@ -14,11 +14,12 @@ Rotated object detection is used to detect rectangular bounding boxes with angle ...@@ -14,11 +14,12 @@ Rotated object detection is used to detect rectangular bounding boxes with angle
## Model Zoo ## Model Zoo
| Model | mAP | Lr Scheduler | Angle | Aug | GPU Number | images/GPU | download | config | | Model | mAP | Lr Scheduler | Angle | Aug | GPU Number | images/GPU | download | config |
|:---:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:| |:---:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:|
| [S2ANet](./s2anet/README.md) | 74.0 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/dota/s2anet_alignconv_2x_dota.yml) | | [S2ANet](./s2anet/README_en.md) | 73.84 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**Notes:** **Notes:**
- if **GPU number** or **mini-batch size** is changed, **learning rate** should be adjusted according to the formula **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)**. - if **GPU number** or **mini-batch size** is changed, **learning rate** should be adjusted according to the formula **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)**.
- Models in model zoo is trained and tested with single scale by default. If `MS` is indicated in the data augmentation column, it means that multi-scale training and multi-scale testing are used. If `RR` is indicated in the data augmentation column, it means that RandomRotate data augmentation is used for training.
## Data Preparation ## Data Preparation
### DOTA Dataset preparation ### DOTA Dataset preparation
...@@ -35,8 +36,16 @@ ${DOTA_ROOT} ...@@ -35,8 +36,16 @@ ${DOTA_ROOT}
└── labelTxt └── labelTxt
``` ```
The image resolution of DOTA dataset is relatively high, so we usually slice the images before training and testing. To slice the images with a single scale, you can use the command below For labeled data, each image corresponds to a txt file with the same name, and each row in the txt file represent a rotated bouding box. The format is as follows:
```
x1 y1 x2 y2 x3 y3 x4 y4 class_name difficult
``` ```
### Slicing data with single scale
The image resolution of DOTA dataset is relatively high, so we usually slice the images before training and testing. To slice the images with a single scale, you can use the command below
``` bash
# slicing labeled data
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \ --input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \
--output_dir ${OUTPUT_DIR}/trainval1024/ \ --output_dir ${OUTPUT_DIR}/trainval1024/ \
...@@ -44,26 +53,37 @@ python configs/rotate/tools/prepare_data.py \ ...@@ -44,26 +53,37 @@ python configs/rotate/tools/prepare_data.py \
--subsize 1024 \ --subsize 1024 \
--gap 200 \ --gap 200 \
--rates 1.0 --rates 1.0
# slicing unlabeled data by setting --image_only
python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/test/ \
--output_dir ${OUTPUT_DIR}/test1024/ \
--coco_json_file DOTA_test1024.json \
--subsize 1024 \
--gap 200 \
--rates 1.0 \
--image_only
``` ```
### Slicing data with multi scale
To slice the images with multiple scales, you can use the command below To slice the images with multiple scales, you can use the command below
``` ``` bash
# slicing labeled data
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \ --input_dirs ${DOTA_ROOT}/train/ ${DOTA_ROOT}/val/ \
--output_dir ${OUTPUT_DIR}/trainval/ \ --output_dir ${OUTPUT_DIR}/trainval/ \
--coco_json_file DOTA_trainval1024.json \ --coco_json_file DOTA_trainval1024.json \
--subsize 1024 \ --subsize 1024 \
--gap 500 \ --gap 500 \
--rates 0.5 1.0 1.5 \ --rates 0.5 1.0 1.5
``` # slicing unlabeled data by setting --image_only
For data without annotations, you should set `--image_only` as follows
```
python configs/rotate/tools/prepare_data.py \ python configs/rotate/tools/prepare_data.py \
--input_dirs ${DOTA_ROOT}/test/ \ --input_dirs ${DOTA_ROOT}/test/ \
--output_dir ${OUTPUT_DIR}/test1024/ \ --output_dir ${OUTPUT_DIR}/test1024/ \
--coco_json_file DOTA_test1024.json \ --coco_json_file DOTA_test1024.json \
--subsize 1024 \ --subsize 1024 \
--gap 200 \ --gap 500 \
--rates 1.0 \ --rates 0.5 1.0 1.5 \
--image_only --image_only
``` ```
......
configs/rotate/s2anet/README.md
浏览文件 @ 38f4a27f
# S2ANet模型 简体中文 | [English](README_en.md)
# S2ANet
## 内容 ## 内容
- [简介](#简介) - [简介](#简介)
- [开始训练](#开始训练)
- [模型库](#模型库) - [模型库](#模型库)
- [使用说明](#使用说明)
- [预测部署](#预测部署) - [预测部署](#预测部署)
- [引用](#引用)
## 简介 ## 简介
[S2ANet](https://arxiv.org/pdf/2008.09397.pdf)是用于检测旋转框的模型，在DOTA 1.0数据集上单尺度训练能达到74.0的mAP. [S2ANet](https://arxiv.org/pdf/2008.09397.pdf)是用于检测旋转框的模型.
## 模型库
| 模型 | Conv类型 | mAP | 学习率策略 | 角度表示 | 数据增广 | GPU数目 | 每GPU图片数目 | 模型下载 | 配置文件 |
|:---:|:------:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:|
| S2ANet | Conv | 71.45 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_conv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_conv_2x_dota.yml) |
| S2ANet | AlignConv | 73.84 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**注意：**
- 如果**GPU卡数**或者**batch size**发生了改变，你需要按照公式 **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)** 调整学习率。
- 模型库中的模型默认使用单尺度训练单尺度测试。如果数据增广一栏标明MS，意味着使用多尺度训练和多尺度测试。如果数据增广一栏标明RR，意味着使用RandomRotate数据增广进行训练。
- 这里使用`multiclass_nms`，与原作者使用nms略有不同。
## 使用说明
## 开始训练 参考[数据准备](../README.md#数据准备)准备数据。
### 1. 训练 ### 1. 训练
...@@ -22,13 +41,13 @@ python tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml ...@@ -22,13 +41,13 @@ python tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml
GPU多卡训练 GPU多卡训练
```bash ```bash
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml
``` ```
可以通过`--eval`开启边训练边测试。 可以通过`--eval`开启边训练边测试。
### 3. 评估 ### 2. 评估
```bash ```bash
python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams
...@@ -36,7 +55,7 @@ python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=out ...@@ -36,7 +55,7 @@ python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=out
python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams
``` ```
### 4. 预测 ### 3. 预测
执行如下命令，会将图像预测结果保存到`output`文件夹下。 执行如下命令，会将图像预测结果保存到`output`文件夹下。
```bash ```bash
python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3 python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3
...@@ -46,37 +65,26 @@ python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=ou ...@@ -46,37 +65,26 @@ python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=ou
python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3 python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3
``` ```
### 5. DOTA数据评估 ### 4. DOTA数据评估
执行如下命令，会在`output`文件夹下将每个图像预测结果保存到同文件夹名的txt文本中。 执行如下命令，会在`output`文件夹下将每个图像预测结果保存到同文件夹名的txt文本中。
``` ```
python tools/infer.py -c configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml -o weights=./weights/s2anet_alignconv_2x_dota.pdparams --infer_dir=/path/to/test/images --output_dir=output --visualize=False --save_results=True python tools/infer.py -c configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams --infer_dir=/path/to/test/images --output_dir=output --visualize=False --save_results=True
``` ```
参考[DOTA Task](https://captain-whu.github.io/DOTA/tasks.html), 评估DOTA数据集需要生成一个包含所有检测结果的zip文件，每一类的检测结果储存在一个txt文件中，txt文件中每行格式为：`image_name score x1 y1 x2 y2 x3 y3 x4 y4`。将生成的zip文件提交到[DOTA Evaluation](https://captain-whu.github.io/DOTA/evaluation.html)的Task1进行评估。你可以执行以下命令生成评估文件 参考[DOTA Task](https://captain-whu.github.io/DOTA/tasks.html), 评估DOTA数据集需要生成一个包含所有检测结果的zip文件，每一类的检测结果储存在一个txt文件中，txt文件中每行格式为：`image_name score x1 y1 x2 y2 x3 y3 x4 y4`。将生成的zip文件提交到[DOTA Evaluation](https://captain-whu.github.io/DOTA/evaluation.html)的Task1进行评估。你可以执行以下命令生成评估文件
``` ```
python configs/rotate/tools/generate_result.py --pred_txt_dir=output/ --output_dir=submit/ --data_type=dota10 python configs/rotate/tools/generate_result.py --pred_txt_dir=output/ --output_dir=submit/ --data_type=dota10
zip -r submit.zip submit zip -r submit.zip submit
``` ```
## 模型库
### S2ANet模型
| 模型 | Conv类型 | mAP | 模型下载 | 配置文件 |
|:-----------:|:----------:|:--------:| :----------:| :---------: |
| S2ANet | Conv | 71.42 | [model](https://paddledet.bj.bcebos.com/models/s2anet_conv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_conv_2x_dota.yml) |
| S2ANet | AlignConv | 74.0 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**注意：** 这里使用`multiclass_nms`，与原作者使用nms略有不同。
## 预测部署 ## 预测部署
Paddle中`multiclass_nms`算子的输入支持四边形输入，因此部署时可以不需要依赖旋转框IOU计算算子。 Paddle中`multiclass_nms`算子的输入支持四边形输入，因此部署时可以不需要依赖旋转框IOU计算算子。
部署教程请参考[预测部署](../../deploy/README.md) 部署教程请参考[预测部署](../../../deploy/README.md)
## Citations ## 引用
``` ```
@article{han2021align, @article{han2021align,
author={J. {Han} and J. {Ding} and J. {Li} and G. -S. {Xia}}, author={J. {Han} and J. {Ding} and J. {Li} and G. -S. {Xia}},
......
configs/rotate/s2anet/README_en.md
浏览文件 @ 38f4a27f
# S2ANet Model English | [简体中文](README.md)
# S2ANet
## Content ## Content
- [S2ANet Model](#s2anet-model) - [Introduction](#Introduction)
- [Content](#content) - [Model Zoo](#Model-Zoo)
- [Introduction](#introduction) - [Getting Start](#Getting-Start)
- [Start Training](#start-training) - [Deployment](#Deployment)
- [1. Train](#1-train) - [Citations](#Citations)
- [2. Evaluation](#2-evaluation)
- [3. Prediction](#3-prediction)
- [4. DOTA Data evaluation](#4-dota-data-evaluation)
- [Model Library](#model-library)
- [S2ANet Model](#s2anet-model-1)
- [Predict Deployment](#predict-deployment)
- [Citations](#citations)
## Introduction ## Introduction
[S2ANet](https://arxiv.org/pdf/2008.09397.pdf) is used to detect rotated objects and acheives 74.0 mAP on DOTA 1.0 dataset. [S2ANet](https://arxiv.org/pdf/2008.09397.pdf) is used to detect rotated objects.
## Model Zoo
| Model | Conv Type | mAP | Lr Scheduler | Angle | Aug | GPU Number | images/GPU | download | config |
|:---:|:------:|:----:|:---------:|:-----:|:--------:|:-----:|:------------:|:-------:|:------:|
| S2ANet | Conv | 71.45 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_conv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_conv_2x_dota.yml) |
| S2ANet | AlignConv | 73.84 | 2x | le135 | - | 4 | 2 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**Notes:**
- if **GPU number** or **mini-batch size** is changed, **learning rate** should be adjusted according to the formula **lr<sub>new</sub> = lr<sub>default</sub> * (batch_size<sub>new</sub> * GPU_number<sub>new</sub>) / (batch_size<sub>default</sub> * GPU_number<sub>default</sub>)**.
- Models in model zoo is trained and tested with single scale by default. If `MS` is indicated in the data augmentation column, it means that multi-scale training and multi-scale testing are used. If `RR` is indicated in the data augmentation column, it means that RandomRotate data augmentation is used for training.
- `multiclass_nms` is used here, which is slightly different from the original author's use of NMS.
## Start Training ## Getting Start
### 2. Train Refer to [Data-Preparation](../README_en.md#Data-Preparation) to prepare data.
### 1. Train
Single GPU Training Single GPU Training
```bash ```bash
...@@ -30,13 +38,13 @@ python tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml ...@@ -30,13 +38,13 @@ python tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml
Multiple GPUs Training Multiple GPUs Training
```bash ```bash
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/rotate/s2anet/s2anet_1x_spine.yml
``` ```
You can use `--eval`to enable train-by-test. You can use `--eval`to enable train-by-test.
### 3. Evaluation ### 2. Evaluation
```bash ```bash
python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams
...@@ -44,7 +52,7 @@ python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=out ...@@ -44,7 +52,7 @@ python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=out
python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams python tools/eval.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams
``` ```
### 4. Prediction ### 3. Prediction
Executing the following command will save the image prediction results to the `output` folder. Executing the following command will save the image prediction results to the `output` folder.
```bash ```bash
python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3 python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=output/s2anet_1x_spine/model_final.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3
...@@ -54,35 +62,24 @@ Prediction using models that provide training: ...@@ -54,35 +62,24 @@ Prediction using models that provide training:
python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3 python tools/infer.py -c configs/rotate/s2anet/s2anet_1x_spine.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_1x_spine.pdparams --infer_img=demo/39006.jpg --draw_threshold=0.3
``` ```
### 5. DOTA Data evaluation ### 4. DOTA Data evaluation
Execute the following command, will save each image prediction result in `output` folder txt text with the same folder name. Execute the following command, will save each image prediction result in `output` folder txt text with the same folder name.
``` ```
python tools/infer.py -c configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml -o weights=./weights/s2anet_alignconv_2x_dota.pdparams --infer_dir=/path/to/test/images --output_dir=output --visualize=False --save_results=True python tools/infer.py -c configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml -o weights=https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams --infer_dir=/path/to/test/images --output_dir=output --visualize=False --save_results=True
``` ```
Refering to [DOTA Task](https://captain-whu.github.io/DOTA/tasks.html), You need to submit a zip file containing results for all test images for evaluation. The detection results of each category are stored in a txt file, each line of which is in the following format Refering to [DOTA Task](https://captain-whu.github.io/DOTA/tasks.html), You need to submit a zip file containing results for all test images for evaluation. The detection results of each category are stored in a txt file, each line of which is in the following format
`image_id score x1 y1 x2 y2 x3 y3 x4 y4`. To evaluate, you should submit the generated zip file to the Task1 of [DOTA Evaluation](https://captain-whu.github.io/DOTA/evaluation.html). You can execute the following command to generate the file `image_id score x1 y1 x2 y2 x3 y3 x4 y4`. To evaluate, you should submit the generated zip file to the Task1 of [DOTA Evaluation](https://captain-whu.github.io/DOTA/evaluation.html). You can execute the following command to generate the file
``` ```
python configs/rotate/tools/generate_result.py --pred_txt_dir=output/ --output_dir=submit/ --data_type=dota10 python configs/rotate/tools/generate_result.py --pred_txt_dir=output/ --output_dir=submit/ --data_type=dota10
zip -r submit.zip submit zip -r submit.zip submit
``` ```
## Model Library ## Deployment
### S2ANet Model
| Model | Conv Type | mAP | Model Download | Configuration File |
|:-----------:|:----------:|:--------:| :----------:| :---------: |
| S2ANet | Conv | 71.42 | [model](https://paddledet.bj.bcebos.com/models/s2anet_conv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_conv_2x_dota.yml) |
| S2ANet | AlignConv | 74.0 | [model](https://paddledet.bj.bcebos.com/models/s2anet_alignconv_2x_dota.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/rotate/s2anet/s2anet_alignconv_2x_dota.yml) |
**Attention:** `multiclass_nms` is used here, which is slightly different from the original author's use of NMS.
## Predict Deployment
The inputs of the `multiclass_nms` operator in Paddle support quadrilateral inputs, so deployment can be done without relying on the rotating frame IOU operator. The inputs of the `multiclass_nms` operator in Paddle support quadrilateral inputs, so deployment can be done without relying on the rotating frame IOU operator.
Please refer to the deployment tutorial[Predict deployment](../../deploy/README_en.md) Please refer to the deployment tutorial[Predict deployment](../../../deploy/README_en.md)
## Citations ## Citations
......
configs/rotate/tools/slicebase.py
浏览文件 @ 38f4a27f
...@@ -222,8 +222,10 @@ class SliceBase(object): ...@@ -222,8 +222,10 @@ class SliceBase(object):
windows = self.get_windows(height, width) windows = self.get_windows(height, width)
self.slice_image_single(resize_img, windows, output_dir, base_name) self.slice_image_single(resize_img, windows, output_dir, base_name)
if not self.image_only: if not self.image_only:
self.slice_anno_single(info['annotation'], windows, output_dir, annos = info['annotation']
base_name) for anno in annos:
anno['poly'] = list(map(lambda x: rate * x, anno['poly']))
self.slice_anno_single(annos, windows, output_dir, base_name)
def check_or_mkdirs(self, path): def check_or_mkdirs(self, path):
if not os.path.exists(path): if not os.path.exists(path):
......
test_tipc/configs/dota/s2anet_alignconv_2x_spine.yml
浏览文件 @ 38f4a27f
_BASE_: [ _BASE_: [
'../../../configs/datasets/spine_coco.yml', '../../../configs/datasets/spine_coco.yml',
'../../../configs/runtime.yml', '../../../configs/runtime.yml',
'../../../configs/dota/_base_/s2anet_optimizer_2x.yml', '../../../configs/rotate/s2anet/_base_/s2anet_optimizer_2x.yml',
'../../../configs/dota/_base_/s2anet.yml', '../../../configs/rotate/s2anet/_base_/s2anet.yml',
'../../../configs/dota/_base_/s2anet_reader.yml', '../../../configs/rotate/s2anet/_base_/s2anet_reader.yml',
] ]
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
......
test_tipc/configs/dota/s2anet_conv_2x_spine.yml
浏览文件 @ 38f4a27f
_BASE_: [ _BASE_: [
'../../../configs/datasets/spine_coco.yml', '../../../configs/datasets/spine_coco.yml',
'../../../configs/runtime.yml', '../../../configs/runtime.yml',
'../../../configs/dota/_base_/s2anet_optimizer_2x.yml', '../../../configs/rotate/s2anet/_base_/s2anet_optimizer_2x.yml',
'../../../configs/dota/_base_/s2anet.yml', '../../../configs/rotate/s2anet/_base_/s2anet.yml',
'../../../configs/dota/_base_/s2anet_reader.yml', '../../../configs/rotate/s2anet/_base_/s2anet_reader.yml',
] ]
weights: output/s2anet_conv_1x_dota/model_final weights: output/s2anet_conv_1x_dota/model_final
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_cos_pretrained.pdparams pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_cos_pretrained.pdparams
......
test_tipc/prepare.sh
浏览文件 @ 38f4a27f
...@@ -152,8 +152,8 @@ else ...@@ -152,8 +152,8 @@ else
cd ./dataset/wider_face/ && tar -xvf wider_tipc.tar && mv -n wider_tipc/* . cd ./dataset/wider_face/ && tar -xvf wider_tipc.tar && mv -n wider_tipc/* .
rm -rf wider_tipc/ && cd ../../ rm -rf wider_tipc/ && cd ../../
# download spine_coco lite data # download spine_coco lite data
wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_tipc.tar --no-check-certificate wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_coco_tipc.tar --no-check-certificate
cd ./dataset/spine_coco/ && tar -xvf spine_tipc.tar && mv -n spine_tipc/* . cd ./dataset/spine_coco/ && tar -xvf spine_coco_tipc.tar && mv -n spine_coco_tipc/* .
rm -rf spine_tipc/ && cd ../../ rm -rf spine_tipc/ && cd ../../
if [[ ${model_name} =~ "s2anet" ]]; then if [[ ${model_name} =~ "s2anet" ]]; then
cd ./ppdet/ext_op && eval "${python} setup.py install" cd ./ppdet/ext_op && eval "${python} setup.py install"
......
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