[MOT] add MOT feature models (#4105)

* add pedestrian vehicle cfgs * fix metric category * fix mot metric * fix mot metric * update pedestrian and vehicle modelzoo * update headtracking modelzoo * add tools for data processing * add data pre-process code * fix tools code format

[MOT] add MOT feature models (#4105)
* add pedestrian vehicle cfgs * fix metric category * fix mot metric * fix mot metric * update pedestrian and vehicle modelzoo * update headtracking modelzoo * add tools for data processing * add data pre-process code * fix tools code format
13c1fef1 · Feng Ni · GitHub · afa7c14d · 13c1fef1 · 13c1fef1
20 changed file
--- a/configs/mot/README.md
+++ b/configs/mot/README.md
@@ -136,7 +136,7 @@ If you use a stronger detection model, you can get better results. Each txt is t
 FairMOT DLA-34 used 2 GPUs for training and mini-batch size as 6 on each GPU, and trained for 30 epoches.


-### FairMOT light model 
+### FairMOT light model
 ### Results on MOT-16 Test Set
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
@@ -158,20 +158,27 @@ If you use a stronger detection model, you can get better results. Each txt is t
 ### FairMOT Results on HT-21 Training Set
 |    backbone      |  input shape |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  download | config |
 | :--------------| :------- | :----: | :----: | :---: | :----: | :---: | :------: | :----: |:----: |
-| DLA-34         | 1088x608 |  67.2 |  70.4  |   9403  |  124840  |  255007  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+| DLA-34         | 1088x608 |  64.7 |  69.0  |   8533  |  148817  |  234970  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |

 ### FairMOT Results on HT-21 Test Set
 |    backbone      |  input shape |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: |:-------: | :----: | :----: |
-| DLA-34         | 1088x608 |  58.2  |  61.3  |  13166   |  141872  |  197074 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
-
-### [Vehicle Tracking](./kitticars/README.md)
-### FairMOT Results on KITTI tracking (2D bounding-boxes) Training Set (Car)
-
-|    backbone    | input shape |  MOTA   |   FPS   |  download | config |
-| :--------------| :------- | :-----: | :-----: | :------: | :----: |
-| DLA-34         | 1088x608 |   53.9  |    -    |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml) |
-
+| DLA-34         | 1088x608 |  60.8  |  62.8  |  12781   |  118109  |  198896 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+
+### [Pedestrian Tracking](./pedestrian/README.md)
+### FairMOT Results on each val-set of Pedestrian category
+|    Dataset      |  input shape |  MOTA  |  IDF1  |  FPS   |  download | config |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  PathTrack    | 1088x608 |  44.9 |    59.3   |    -   |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml) |
+|  VisDrone     | 1088x608 |  49.2 |   63.1 |    -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_pedestrian.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml) |
+
+### [Vehicle Tracking](./vehicle/README.md)
+### FairMOT Results on each val-set of Vehicle category
+|    Dataset      |  input shape |  MOTA  |  IDF1  |  FPS   |  download | config |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  BDD100K      | 1088x608 |  34.9 |  39.9  |    -    | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100k_vehicle.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml) |
+|  KITTI        | 1088x608 |  53.9 |    -   |    -   |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitti_vehicle.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_kitti_vehicle.yml) |
+|  VisDrone     | 1088x608 |  29.8 |   51.3 |    -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_vehicle.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_visdrone_vehicle.yml) |

 ## Dataset Preparation


--- a/configs/mot/README_cn.md
+++ b/configs/mot/README_cn.md
@@ -154,23 +154,31 @@ wget https://dataset.bj.bcebos.com/mot/det_results_dir.zip
 ## 特色垂类跟踪模型

 ### [人头跟踪（Head Tracking)](./headtracking21/README.md)
-
 ### FairMOT在HT-21 Training Set上结果
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  下载链接 | 配置文件 |
 | :--------------| :------- | :----: | :----: | :---: | :----: | :---: | :------: | :----: |:----: |
-| DLA-34         | 1088x608 |  67.2 |  70.4  |   9403  |  124840  |  255007  |     -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+| DLA-34         | 1088x608 |  64.7 |  69.0  |   8533  |  148817  |  234970  |     -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |

 ### FairMOT在HT-21 Test Set上结果
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |   IDS  |   FP   |   FN   |    FPS   |  下载链接  | 配置文件 |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: |:-------: | :----: | :----: |
-| DLA-34         | 1088x608 |  58.2  |  61.3  |  13166   |  141872  |  197074 |    -     | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
-
-### [车辆跟踪 (Vehicle Tracking)](./kitticars/README.md)
-### FairMOT在KITTI tracking (2D bounding-boxes) Training Set上Car类别的结果
-
-|    骨干网络      |  输入尺寸 |  MOTA   |   FPS   |  下载链接 | 配置文件 |
-| :--------------| :------- | :-----: | :-----: | :------: | :----: |
-| DLA-34         | 1088x608 |   53.9  |    -    |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml) |
+| DLA-34         | 1088x608 |  60.8  |  62.8  |  12781   |  118109  |  198896 |    -     | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+
+### [行人跟踪 (Pedestrian Tracking)](./pedestrian/README.md)
+### FairMOT在各个数据集val-set上Pedestrian类别的结果
+|    数据集      |  输入尺寸 |  MOTA  |  IDF1  |  FPS   |  下载链接 | 配置文件 |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  PathTrack    | 1088x608 |  44.9 |    59.3   |    -   |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml) |
+|  VisDrone     | 1088x608 |  49.2 |   63.1 |    -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_pedestrian.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml) |
+
+### [车辆跟踪 (Vehicle Tracking)](./vehicle/README.md)
+### FairMOT在各个数据集val-set上Vehicle类别的结果
+
+|    数据集      |  输入尺寸 |  MOTA  |  IDF1  |  FPS   |  下载链接 | 配置文件 |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  BDD100K      | 1088x608 |  34.9 |  39.9  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100k_vehicle.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml) |
+|  KITTI        | 1088x608 |  53.9 |    -   |    -   |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitti_vehicle.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_kitti_vehicle.yml) |
+|  VisDrone     | 1088x608 |  29.8 |   51.3 |    -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_vehicle.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/vehicle/fairmot_dla34_30e_1088x608_visdrone_vehicle.yml) |


 ## 数据集准备

--- a/configs/mot/fairmot/README.md
+++ b/configs/mot/fairmot/README.md
@@ -36,7 +36,7 @@ English | [简体中文](README_cn.md)
 **Notes:**
 FairMOT DLA-34 used 2 GPUs for training and mini-batch size as 6 on each GPU, and trained for 30 epoches.

-### FairMOT light model 
+### FairMOT light model
 ### Results on MOT-16 Test Set
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |

--- a/configs/mot/headtracking21/README_cn.md
+++ b/configs/mot/headtracking21/README_cn.md
@@ -13,15 +13,15 @@
 ### FairMOT在HT-21 Training Set上结果
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  下载链接 | 配置文件 |
 | :--------------| :------- | :----: | :----: | :---: | :----: | :---: | :------: | :----: |:----: |
-| DLA-34         | 1088x608 |  67.2 |  70.4  |   9403  |  124840  |  255007  |     -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+| DLA-34         | 1088x608 |  64.7 |  69.0  |   8533  |  148817  |  234970  |     -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_headtracking21.yml) |

 ### FairMOT在HT-21 Test Set上结果
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |   IDS  |   FP   |   FN   |    FPS   |  下载链接  | 配置文件 |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: |:-------: | :----: | :----: |
-| DLA-34         | 1088x608 |  58.2  |  61.3  |  13166   |  141872  |  197074 |    -     | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/headtracking21/fairmot_dla34_30e_1088x608_headtracking21.yml) |
+| DLA-34         | 1088x608 |  60.8  |  62.8  |  12781   |  118109  |  198896 |    -     | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_headtracking21.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_headtracking21.yml) |

 **注意:**
- FairMOT使用2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。
+ FairMOT DLA-34使用2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。目前MOTA精度位于MOT官网[Head Tracking 21](https://motchallenge.net/results/Head_Tracking_21)榜单榜首。

 ## 快速开始


--- a/configs/mot/kitticars/README_cn.md
+++ b/configs/mot/kitticars/README_cn.md
-[English](README.md) | 简体中文
-# 特色垂类跟踪模型
-
-## 车辆跟踪 (Vehicle Tracking)
-
-车辆跟踪的主要应用之一是交通监控。
-[KITTI-Tracking](http://www.cvlibs.net/datasets/kitti/eval_tracking.php)是一个包含市区、乡村和高速公路等场景采集的数据集，每张图像中最多达15辆车和30个行人，还有各种程度的遮挡与截断。其中用于目标跟踪的数据集一共有50个视频序列，21个为训练集，29个为测试集，目标是估计类别“Car”和”Pedestrian“的目标轨迹，此处只使用类别“Car”。
-<div align="center">
-  <img src="../../../docs/images/kitticars_fairmot.gif" width='800'/>
-</div>
-
-## 模型库
-
-### FairMOT在KITTI tracking (2D bounding-boxes) Training Set上Car类别的结果
-
-|    骨干网络      |  输入尺寸 |  MOTA   |   FPS   |  下载链接 | 配置文件 |
-| :--------------| :------- | :-----: | :-----: | :------: | :----: |
-| DLA-34         | 1088x608 |   53.9  |    -    |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml) |
-
-**注意:**
- FairMOT使用2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。
-
-## 快速开始
-
-### 1. 训练
-使用2个GPU通过如下命令一键式启动训练
-```bash
-python -m paddle.distributed.launch --log_dir=./fairmot_dla34_30e_1088x608_kitticars/ --gpus 0,1 tools/train.py -c configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml
-```
-
-### 2. 评估
-使用单张GPU通过如下命令一键式启动评估
-```bash
-# 使用PaddleDetection发布的权重
-CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams
-
-# 使用训练保存的checkpoint
-CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml -o weights=output/fairmot_dla34_30e_1088x608_kitticars/model_final.pdparams
-```
-
-### 3. 预测
-使用单个GPU通过如下命令预测一个视频，并保存为视频
-```bash
-# 预测一个视频
-CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams --video_file={your video name}.mp4  --save_videos
-```
-**注意:**
- 请先确保已经安装了[ffmpeg](https://ffmpeg.org/ffmpeg.html), Linux(Ubuntu)平台可以直接用以下命令安装：`apt-get update && apt-get install -y ffmpeg`。
-
-### 4. 导出预测模型
-```bash
-CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitticars.pdparams
-```
-
-### 5. 用导出的模型基于Python去预测
-```bash
-python deploy/python/mot_jde_infer.py --model_dir=output_inference/fairmot_dla34_30e_1088x608_kitticars --video_file={your video name}.mp4 --device=GPU --save_mot_txts
-```
-**注意:**
- 跟踪模型是对视频进行预测，不支持单张图的预测，默认保存跟踪结果可视化后的视频，可添加`--save_mot_txts`表示保存跟踪结果的txt文件，或`--save_images`表示保存跟踪结果可视化图片。
-
-## 引用
-```
-@article{zhang2020fair,
-  title={FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking},
-  author={Zhang, Yifu and Wang, Chunyu and Wang, Xinggang and Zeng, Wenjun and Liu, Wenyu},
-  journal={arXiv preprint arXiv:2004.01888},
-  year={2020}
-}
-@INPROCEEDINGS{Geiger2012CVPR,
-  author = {Andreas Geiger and Philip Lenz and Raquel Urtasun},
-  title = {Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},
-  booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
-  year = {2012}
-}
-```
--- a/configs/mot/kitticars/README.md
+++ b/configs/mot/kitticars/README.md
--- a/configs/mot/pedestrian/README_cn.md
+++ b/configs/mot/pedestrian/README_cn.md
+[English](README.md) | 简体中文
+# 特色垂类跟踪模型
+
+## 大规模行人跟踪 (Pedestrian Tracking)
+
+行人跟踪的主要应用之一是交通监控。
+
+[PathTrack](https://www.trace.ethz.ch/publications/2017/pathtrack/index.html)包含720个视频序列，有着超过15000个行人的轨迹。包含了街景、舞蹈、体育运动、采访等各种场景的，大部分是移动摄像头拍摄场景。该数据集只有Pedestrian一类标注作为跟踪任务。
+
+[VisDrone](http://aiskyeye.com)是无人机视角拍摄的数据集，是以俯视视角为主。该数据集涵盖不同位置（取自中国数千个相距数千公里的14个不同城市）、不同环境（城市和乡村）、不同物体（行人、车辆、自行车等）和不同密度（稀疏和拥挤的场景）。[VisDrone2019-MOT](https://github.com/VisDrone/VisDrone-Dataset)包含56个视频序列用于训练，7个视频序列用于验证。此处针对VisDrone2019-MOT多目标跟踪数据集进行提取，抽取出类别为pedestrian和people的数据组合成一个大的Pedestrian类别。
+
+
+## 模型库
+
+### FairMOT在各个数据集val-set上Pedestrian类别的结果
+
+|    数据集      |  输入尺寸 |  MOTA  |  IDF1  |  FPS   |  下载链接 | 配置文件 |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  PathTrack    | 1088x608 |  44.9 |    59.3   |    -   |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_pathtrack.yml) |
+|  VisDrone     | 1088x608 |  49.2 |   63.1 |    -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_pedestrian.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml) |
+
+**注意:**
+ FairMOT均使用DLA-34为骨干网络，2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。
+
+
+## 数据集准备和处理
+
+### 1、数据集处理代码说明
+代码统一都在tools目录下
+```
+# visdrone
+tools/visdrone/visdrone2mot.py: 生成visdrone_pedestrian据集
+```
+
+### 2、visdrone_pedestrian数据集处理
+```
+# 复制tool/visdrone/visdrone2mot.py到数据集目录下
+# 生成visdrone_pedestrian MOT格式的数据，抽取类别classes=1,2 (pedestrian, people)
+<<--生成前目录-->>
+├── VisDrone2019-MOT-val
+│   ├── annotations
+│   ├── sequences
+│   ├── visdrone2mot.py
+<<--生成后目录-->>
+├── VisDrone2019-MOT-val
+│   ├── annotations
+│   ├── sequences
+│   ├── visdrone2mot.py
+│   ├── visdrone_pedestrian
+│   │   ├── images
+│   │   │   ├── val
+│   │   ├── labels_with_ids
+│   │   │   ├── val
+# 执行
+python visdrone2mot.py --transMot=True --data_name=visdrone_pedestrian --phase=val
+python visdrone2mot.py --transMot=True --data_name=visdrone_pedestrian --phase=train
+```
+
+## 快速开始
+
+### 1. 训练
+使用2个GPU通过如下命令一键式启动训练
+```bash
+python -m paddle.distributed.launch --log_dir=./fairmot_dla34_30e_1088x608_pathtrack/ --gpus 0,1 tools/train.py -c configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml
+```
+
+### 2. 评估
+使用单张GPU通过如下命令一键式启动评估
+```bash
+# 使用PaddleDetection发布的权重
+CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams
+
+# 使用训练保存的checkpoint
+CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml -o weights=output/fairmot_dla34_30e_1088x608_pathtrack/model_final.pdparams
+```
+
+### 3. 预测
+使用单个GPU通过如下命令预测一个视频，并保存为视频
+```bash
+# 预测一个视频
+CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams --video_file={your video name}.mp4  --save_videos
+```
+**注意:**
+ 请先确保已经安装了[ffmpeg](https://ffmpeg.org/ffmpeg.html), Linux(Ubuntu)平台可以直接用以下命令安装：`apt-get update && apt-get install -y ffmpeg`。
+
+### 4. 导出预测模型
+```bash
+CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_pathtrack.pdparams
+```
+
+### 5. 用导出的模型基于Python去预测
+```bash
+python deploy/python/mot_jde_infer.py --model_dir=output_inference/fairmot_dla34_30e_1088x608_pathtrack --video_file={your video name}.mp4 --device=GPU --save_mot_txts
+```
+**注意:**
+ 跟踪模型是对视频进行预测，不支持单张图的预测，默认保存跟踪结果可视化后的视频，可添加`--save_mot_txts`表示保存跟踪结果的txt文件，或`--save_images`表示保存跟踪结果可视化图片。
+
+## 引用
+```
+@article{zhang2020fair,
+  title={FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking},
+  author={Zhang, Yifu and Wang, Chunyu and Wang, Xinggang and Zeng, Wenjun and Liu, Wenyu},
+  journal={arXiv preprint arXiv:2004.01888},
+  year={2020}
+}
+
+@INPROCEEDINGS{8237302,
+author={S. {Manen} and M. {Gygli} and D. {Dai} and L. V. {Gool}},
+booktitle={2017 IEEE International Conference on Computer Vision (ICCV)},
+title={PathTrack: Fast Trajectory Annotation with Path Supervision},
+year={2017},
+volume={},
+number={},
+pages={290-299},
+doi={10.1109/ICCV.2017.40},
+ISSN={2380-7504},
+month={Oct},}
+
+@article{zhu2018vision,
+  title={Vision meets drones: A challenge},
+  author={Zhu, Pengfei and Wen, Longyin and Bian, Xiao and Ling, Haibin and Hu, Qinghua},
+  journal={arXiv preprint arXiv:1804.07437},
+  year={2018}
+}
+@article{zhu2020vision,
+  title={Vision Meets Drones: Past, Present and Future},
+  author={Zhu, Pengfei and Wen, Longyin and Du, Dawei and Bian, Xiao and Hu, Qinghua and Ling, Haibin},
+  journal={arXiv preprint arXiv:2001.06303},
+  year={2020}
+}
+```
--- a/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml
+++ b/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_pathtrack.yml
+_BASE_: [
+  '../fairmot/fairmot_dla34_30e_1088x608.yml'
+]
+
+weights: output/fairmot_dla34_30e_1088x608_pathtrack/model_final
+
+# for MOT training
+TrainDataset:
+  !MOTDataSet
+    dataset_dir: dataset/mot
+    image_lists: ['pathtrack.train']
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']
+
+# for MOT evaluation
+# If you want to change the MOT evaluation dataset, please modify 'data_root'
+EvalMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    data_root: pathtrack/images/test
+    keep_ori_im: False # set True if save visualization images or video, or used in DeepSORT
+
+# for MOT video inference
+TestMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    keep_ori_im: True # set True if save visualization images or video
--- a/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml
+++ b/configs/mot/pedestrian/fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml
+_BASE_: [
+  '../fairmot/fairmot_dla34_30e_1088x608.yml'
+]
+
+weights: output/fairmot_dla34_30e_1088x608_visdrone_pedestrian/model_final
+
+# for MOT training
+TrainDataset:
+  !MOTDataSet
+    dataset_dir: dataset/mot
+    image_lists: ['visdrone_pedestrian.train']
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']
+
+# for MOT evaluation
+# If you want to change the MOT evaluation dataset, please modify 'data_root'
+EvalMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    data_root: visdrone_pedestrian/images/val
+    keep_ori_im: False # set True if save visualization images or video, or used in DeepSORT
+
+# for MOT video inference
+TestMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    keep_ori_im: True # set True if save visualization images or video
--- a/configs/mot/pedestrian/tools/visdrone/visdrone2mot.py
+++ b/configs/mot/pedestrian/tools/visdrone/visdrone2mot.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import glob
+import os
+import os.path as osp
+import cv2
+import argparse
+import numpy as np
+import random
+
+# The object category indicates the type of annotated object, 
+# (i.e., ignored regions(0), pedestrian(1), people(2), bicycle(3), car(4), van(5), truck(6), tricycle(7), awning-tricycle(8), bus(9), motor(10),others(11))
+
+
+def mkdir_if_missing(d):
+    if not osp.exists(d):
+        os.makedirs(d)
+
+
+def genGtFile(seqPath, outPath, classes=[]):
+    id_idx = 0
+    old_idx = -1
+    with open(seqPath, 'r') as singleSeqFile:
+        motLine = []
+        allLines = singleSeqFile.readlines()
+        for line in allLines:
+            line = line.replace('\n', '')
+            line = line.split(',')
+            # exclude occlusion!='2'
+            if line[-1] != '2' and line[7] in classes:
+                if old_idx != int(line[1]):
+                    id_idx += 1
+                    old_idx = int(line[1])
+                newLine = line[0:6]
+                newLine[1] = str(id_idx)
+                newLine.append('1')
+                if (len(classes) > 1):
+                    class_index = str(classes.index(line[7]) + 1)
+                    newLine.append(class_index)
+                else:
+                    newLine.append('1')  # use permenant class '1'
+                newLine.append('1')
+                motLine.append(newLine)
+        mkdir_if_missing(outPath)
+        gtFilePath = osp.join(outPath, 'gt.txt')
+        with open(gtFilePath, 'w') as gtFile:
+            motLine = list(map(lambda x: str.join(',', x), motLine))
+            motLineStr = str.join('\n', motLine)
+            gtFile.write(motLineStr)
+
+
+def genSeqInfo(img1Path, seqName):
+    imgPaths = glob.glob(img1Path + '/*.jpg')
+    seqLength = len(imgPaths)
+    if seqLength > 0:
+        image1 = cv2.imread(imgPaths[0])
+        imgHeight = image1.shape[0]
+        imgWidth = image1.shape[1]
+    else:
+        imgHeight = 0
+        imgWidth = 0
+    seqInfoStr = f'''[Sequence]\nname={seqName}\nimDir=img1\nframeRate=30\nseqLength={seqLength}\nimWidth={imgWidth}\nimHeight={imgHeight}\nimExt=.jpg'''
+    seqInfoPath = img1Path.replace('/img1', '')
+    with open(seqInfoPath + '/seqinfo.ini', 'w') as seqFile:
+        seqFile.write(seqInfoStr)
+
+
+def copyImg(img1Path, gtTxtPath, outputFileName):
+    with open(gtTxtPath, 'r') as gtFile:
+        allLines = gtFile.readlines()
+        imgList = []
+        for line in allLines:
+            imgIdx = int(line.split(',')[0])
+            if imgIdx not in imgList:
+                imgList.append(imgIdx)
+                seqName = gtTxtPath.replace('./{}/'.format(outputFileName),
+                                            '').replace('/gt/gt.txt', '')
+                sourceImgPath = osp.join('./sequences', seqName,
+                                         '{:07d}.jpg'.format(imgIdx))
+                os.system(f'cp {sourceImgPath} {img1Path}')
+
+
+def genMotLabels(datasetPath, outputFileName, classes=['2']):
+    mkdir_if_missing(osp.join(datasetPath, outputFileName))
+    annotationsPath = osp.join(datasetPath, 'annotations')
+    annotationsList = glob.glob(osp.join(annotationsPath, '*.txt'))
+    for annotationPath in annotationsList:
+        seqName = annotationPath.split('/')[-1].replace('.txt', '')
+        mkdir_if_missing(osp.join(datasetPath, outputFileName, seqName, 'gt'))
+        mkdir_if_missing(osp.join(datasetPath, outputFileName, seqName, 'img1'))
+        genGtFile(annotationPath,
+                  osp.join(datasetPath, outputFileName, seqName, 'gt'), classes)
+        img1Path = osp.join(datasetPath, outputFileName, seqName, 'img1')
+        gtTxtPath = osp.join(datasetPath, outputFileName, seqName, 'gt/gt.txt')
+        copyImg(img1Path, gtTxtPath, outputFileName)
+        genSeqInfo(img1Path, seqName)
+
+
+def deleteFileWhichImg1IsEmpty(mot16Path, dataType='train'):
+    path = mot16Path
+    data_images_train = osp.join(path, 'images', f'{dataType}')
+    data_images_train_seqs = glob.glob(data_images_train + '/*')
+    if (len(data_images_train_seqs) == 0):
+        print('dataset is empty!')
+    for data_images_train_seq in data_images_train_seqs:
+        data_images_train_seq_img1 = osp.join(data_images_train_seq, 'img1')
+        if len(glob.glob(data_images_train_seq_img1 + '/*.jpg')) == 0:
+            print(f"os.system(rm -rf {data_images_train_seq})")
+            os.system(f'rm -rf {data_images_train_seq}')
+
+
+def formatMot16Path(dataPath, pathType='train'):
+    train_path = osp.join(dataPath, 'images', pathType)
+    mkdir_if_missing(train_path)
+    os.system(f'mv {dataPath}/* {train_path}')
+
+
+def VisualGt(dataPath, phase='train'):
+    seqList = sorted(glob.glob(osp.join(dataPath, 'images', phase) + '/*'))
+    seqIndex = random.randint(0, len(seqList) - 1)
+    seqPath = seqList[seqIndex]
+    gt_path = osp.join(seqPath, 'gt', 'gt.txt')
+    img_list_path = sorted(glob.glob(osp.join(seqPath, 'img1', '*.jpg')))
+    imgIndex = random.randint(0, len(img_list_path))
+    img_Path = img_list_path[imgIndex]
+    frame_value = int(img_Path.split('/')[-1].replace('.jpg', ''))
+    gt_value = np.loadtxt(gt_path, dtype=int, delimiter=',')
+    gt_value = gt_value[gt_value[:, 0] == frame_value]
+    get_list = gt_value.tolist()
+    img = cv2.imread(img_Path)
+    colors = [[255, 0, 0], [255, 255, 0], [255, 0, 255], [0, 255, 0],
+              [0, 255, 255], [0, 0, 255]]
+    for seq, _id, pl, pt, w, h, _, bbox_class, _ in get_list:
+        cv2.putText(img,
+                    str(bbox_class), (pl, pt), cv2.FONT_HERSHEY_PLAIN, 2,
+                    colors[bbox_class - 1])
+        cv2.rectangle(
+            img, (pl, pt), (pl + w, pt + h),
+            colors[bbox_class - 1],
+            thickness=2)
+    cv2.imwrite('testGt.jpg', img)
+
+
+def VisualDataset(datasetPath, phase='train', seqName='', frameId=1):
+    trainPath = osp.join(datasetPath, 'labels_with_ids', phase)
+    seq1Paths = osp.join(trainPath, seqName)
+    seq_img1_path = osp.join(seq1Paths, 'img1')
+    label_with_idPath = osp.join(seq_img1_path, '%07d' % frameId) + '.txt'
+    image_path = label_with_idPath.replace('labels_with_ids', 'images').replace(
+        '.txt', '.jpg')
+    seqInfoPath = str.join('/', image_path.split('/')[:-2])
+    seqInfoPath = seqInfoPath + '/seqinfo.ini'
+    seq_info = open(seqInfoPath).read()
+    width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+        '\nimHeight')])
+    height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+        '\nimExt')])
+
+    with open(label_with_idPath, 'r') as label:
+        allLines = label.readlines()
+        images = cv2.imread(image_path)
+        for line in allLines:
+            line = line.split(' ')
+            line = list(map(lambda x: float(x), line))
+            c1, c2, w, h = line[2:6]
+            x1 = c1 - w / 2
+            x2 = c2 - h / 2
+            x3 = c1 + w / 2
+            x4 = c2 + h / 2
+            cv2.rectangle(
+                images, (int(x1 * width), int(x2 * height)),
+                (int(x3 * width), int(x4 * height)), (255, 0, 0),
+                thickness=2)
+        cv2.imwrite('test.jpg', images)
+
+
+def gen_image_list(dataPath, datType):
+    inputPath = f'{dataPath}/images/{datType}'
+    pathList = glob.glob(inputPath + '/*')
+    pathList = sorted(pathList)
+    allImageList = []
+    for pathSingle in pathList:
+        imgList = sorted(glob.glob(osp.join(pathSingle, 'img1', '*.jpg')))
+        for imgPath in imgList:
+            allImageList.append(imgPath)
+    with open(f'{dataPath}.{datType}', 'w') as image_list_file:
+        allImageListStr = str.join('\n', allImageList)
+        image_list_file.write(allImageListStr)
+
+
+def gen_labels_mot(MOT_data, phase='train'):
+    seq_root = './{}/images/{}'.format(MOT_data, phase)
+    label_root = './{}/labels_with_ids/{}'.format(MOT_data, phase)
+    mkdir_if_missing(label_root)
+    seqs = [s for s in os.listdir(seq_root)]
+    print('seqs => ', seqs)
+    tid_curr = 0
+    tid_last = -1
+    for seq in seqs:
+        seq_info = open(osp.join(seq_root, seq, 'seqinfo.ini')).read()
+        seq_width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+            '\nimHeight')])
+        seq_height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+            '\nimExt')])
+
+        gt_txt = osp.join(seq_root, seq, 'gt', 'gt.txt')
+        gt = np.loadtxt(gt_txt, dtype=np.float64, delimiter=',')
+
+        seq_label_root = osp.join(label_root, seq, 'img1')
+        mkdir_if_missing(seq_label_root)
+
+        for fid, tid, x, y, w, h, mark, label, _ in gt:
+            # if mark == 0 or not label == 1: 
+            #     continue
+            fid = int(fid)
+            tid = int(tid)
+            if not tid == tid_last:
+                tid_curr += 1
+                tid_last = tid
+            x += w / 2
+            y += h / 2
+            label_fpath = osp.join(seq_label_root, '{:07d}.txt'.format(fid))
+            label_str = '0 {:d} {:.6f} {:.6f} {:.6f} {:.6f}\n'.format(
+                tid_curr, x / seq_width, y / seq_height, w / seq_width,
+                h / seq_height)
+            with open(label_fpath, 'a') as f:
+                f.write(label_str)
+
+
+def parse_arguments():
+    parser = argparse.ArgumentParser(description='input method')
+    parser.add_argument("--transMot", type=bool, default=False)
+    parser.add_argument("--genMot", type=bool, default=False)
+    parser.add_argument("--formatMotPath", type=bool, default=False)
+    parser.add_argument("--deleteEmpty", type=bool, default=False)
+    parser.add_argument("--genLabelsMot", type=bool, default=False)
+    parser.add_argument("--genImageList", type=bool, default=False)
+    parser.add_argument("--visualImg", type=bool, default=False)
+    parser.add_argument("--visualGt", type=bool, default=False)
+    parser.add_argument("--data_name", type=str, default='visdrone_pedestrain')
+    parser.add_argument("--phase", type=str, default='train')
+    parser.add_argument(
+        "--classes", type=str, default='1, 2')  # pedestrian and people
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_arguments()
+    classes = args.classes.split(',')
+    datasetPath = './'
+    dataName = args.data_name
+    phase = args.phase
+    if args.transMot:
+        genMotLabels(datasetPath, dataName, classes)
+        formatMot16Path(dataName, pathType=phase)
+        mot16Path = f'./{dataName}'
+        deleteFileWhichImg1IsEmpty(mot16Path, dataType=phase)
+        gen_labels_mot(dataName, phase=phase)
+        gen_image_list(dataName, phase)
+    if args.genMot:
+        genMotLabels(datasetPath, dataName, classes)
+    if args.formatMotPath:
+        formatMot16Path(dataName, pathType=phase)
+    if args.deleteEmpty:
+        mot16Path = f'./{dataName}'
+        deleteFileWhichImg1IsEmpty(mot16Path, dataType=phase)
+    if args.genLabelsMot:
+        gen_labels_mot(dataName, phase=phase)
+    if args.genImageList:
+        gen_image_list(dataName, phase)
+    if args.visualGt:
+        VisualGt(f'./{dataName}', phase)
+    if args.visualImg:
+        seqName = 'uav0000137_00458_v'
+        frameId = 43
+        VisualDataset(
+            f'./{dataName}', phase=phase, seqName=seqName, frameId=frameId)
--- a/configs/mot/vehicle/README.md
+++ b/configs/mot/vehicle/README.md
+README_cn.md
\ No newline at end of file
--- a/configs/mot/vehicle/README_cn.md
+++ b/configs/mot/vehicle/README_cn.md
+[English](README.md) | 简体中文
+# 特色垂类跟踪模型
+
+## 车辆跟踪 (Vehicle Tracking)
+
+车辆跟踪的主要应用之一是交通监控。在监控场景中，大多是从公共区域的监控摄像头视角拍摄车辆，获取图像后再进行车辆检测和跟踪。
+
+
+[BDD100K](https://www.bdd100k.com)是伯克利大学AI实验室（BAIR）提出的一个驾驶视频数据集，是以驾驶员视角为主。该数据集不仅分多类别标注，还分晴天、多云等六种天气，住宅区、公路等六种场景，白天、夜晚等三个时间段，以及是否遮挡、是否截断。BDD100K MOT数据集包含1400个视频序列用于训练，200个视频序列用于验证。每个视频序列大约40秒长，每秒5帧，因此每个视频大约有200帧。此处针对BDD100K MOT数据集进行提取，抽取出类别为car, truck, bus, trailer, other vehicle的数据组合成一个Vehicle类别。
+
+[KITTI](http://www.cvlibs.net/datasets/kitti)是一个包含市区、乡村和高速公路等场景采集的数据集，每张图像中最多达15辆车和30个行人，还有各种程度的遮挡与截断。[KITTI-Tracking](http://www.cvlibs.net/datasets/kitti/eval_tracking.php)(2D bounding-boxes)数据集一共有50个视频序列，21个为训练集，29个为测试集，目标是估计类别Car和Pedestrian的目标轨迹，此处抽取出类别为Car的数据作为一个Vehicle类别。
+
+[VisDrone](http://aiskyeye.com)是无人机视角拍摄的数据集，是以俯视视角为主。该数据集涵盖不同位置（取自中国数千个相距数千公里的14个不同城市）、不同环境（城市和乡村）、不同物体（行人、车辆、自行车等）和不同密度（稀疏和拥挤的场景）。[VisDrone2019-MOT](https://github.com/VisDrone/VisDrone-Dataset)包含56个视频序列用于训练，7个视频序列用于验证。此处针对VisDrone2019-MOT多目标跟踪数据集进行提取，抽取出类别为car、van、truck、bus的数据组合成一个Vehicle类别。
+
+
+## 模型库
+
+### FairMOT在各个数据集val-set上Vehicle类别的结果
+
+|    数据集      |  输入尺寸 |  MOTA  |  IDF1  |  FPS   |  下载链接 | 配置文件 |
+| :-------------| :------- | :----: | :----: | :----: | :-----: |:------: |
+|  BDD100K      | 1088x608 |  34.9 |  39.9  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml) |
+|  KITTI        | 1088x608 |  53.9 |    -   |    -   |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_kitti_vehicle.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_kitti_vehicle.yml) |
+|  VisDrone     | 1088x608 |  29.8 |   51.3 |    -   | [下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_visdrone_vehicle.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608_visdrone_vehicle.yml) |
+
+**注意:**
+ FairMOT均使用DLA-34为骨干网络，2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。
+
+
+## 数据集准备和处理
+
+### 1、数据集处理代码说明
+代码统一都在tools目录下
+```
+# bdd100kmot
+tools/bdd100kmot/gen_bdd100kmot_vehicle.sh：通过执行bdd100k2mot.py和gen_labels_MOT.py生成bdd100kmot_vehicle 数据集
+tools/bdd100kmot/bdd100k2mot.py：将bdd100k全集转换成mot格式
+tools/bdd100kmot/gen_labels_MOT.py：生成单类别的labels_with_ids文件
+# visdrone
+tools/visdrone/visdrone2mot.py：生成visdrone_vehicle
+```
+
+### 2、bdd100k_vehicle数据集处理
+```
+# 复制tools/bdd100kmot里的代码到数据集目录下
+# 生成bdd100kmot_vehicle MOT格式的数据，抽取类别classes=2,3,4,9,10 (car, truck, bus, trailer, other vehicle)
+<<--生成前目录-->>
+├── bdd100k_path
+│   ├── images
+│   ├── labels
+<<--生成后目录-->>
+├── bdd100k_path
+│   ├── images
+│   ├── labels
+│   ├── bdd100kmot_vehicle
+│   │   ├── images
+│   │   │   ├── train
+│   │   │   ├── val
+│   │   ├── labels_with_ids
+│   │   │   ├── train
+│   │   │   ├── val
+# 执行
+sh gen_bdd100kmot_vehicle.sh
+```
+
+### 3、visdrone_vehicle数据集处理
+```
+# 复制tools/visdrone/visdrone2mot.py到数据集目录下
+# 生成visdrone_vehicle MOT格式的数据，抽取类别classes=4,5,6,9 (car, van, truck, bus)
+<<--生成前目录-->>
+├── VisDrone2019-MOT-val
+│   ├── annotations
+│   ├── sequences
+│   ├── visdrone2mot.py
+<<--生成后目录-->>
+├── VisDrone2019-MOT-val
+│   ├── annotations
+│   ├── sequences
+│   ├── visdrone2mot.py
+│   ├── visdrone-vehicle
+│   │   ├── images
+│   │   │   ├── val
+│   │   ├── labels_with_ids
+│   │   │   ├── val
+# 执行
+python visdrone2mot.py --transMot=True --data_name=visdrone_vehicle --phase=val
+python visdrone2mot.py --transMot=True --data_name=visdrone_vehicle --phase=train
+```
+
+## 快速开始
+
+### 1. 训练
+使用2个GPU通过如下命令一键式启动训练
+```bash
+python -m paddle.distributed.launch --log_dir=./fairmot_dla34_30e_1088x608_bdd100k_vehicle/ --gpus 0,1 tools/train.py -c configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml
+```
+
+### 2. 评估
+使用单张GPU通过如下命令一键式启动评估
+```bash
+# 使用PaddleDetection发布的权重
+CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100k_vehicle.pdparams
+
+# 使用训练保存的checkpoint
+CUDA_VISIBLE_DEVICES=0 python tools/eval_mot.py -c configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml -o weights=output/fairmot_dla34_30e_1088x608_bdd100k_vehicle/model_final.pdparams
+```
+
+### 3. 预测
+使用单个GPU通过如下命令预测一个视频，并保存为视频
+```bash
+# 预测一个视频
+CUDA_VISIBLE_DEVICES=0 python tools/infer_mot.py -c configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100k_vehicle.pdparams --video_file={your video name}.mp4  --save_videos
+```
+**注意:**
+ 请先确保已经安装了[ffmpeg](https://ffmpeg.org/ffmpeg.html), Linux(Ubuntu)平台可以直接用以下命令安装：`apt-get update && apt-get install -y ffmpeg`。
+
+### 4. 导出预测模型
+```bash
+CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100k_vehicle.yml -o weights=https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608_bdd100k_vehicle.pdparams
+```
+
+### 5. 用导出的模型基于Python去预测
+```bash
+python deploy/python/mot_jde_infer.py --model_dir=output_inference/fairmot_dla34_30e_1088x608_bdd100k_vehicle --video_file={your video name}.mp4 --device=GPU --save_mot_txts
+```
+**注意:**
+ 跟踪模型是对视频进行预测，不支持单张图的预测，默认保存跟踪结果可视化后的视频，可添加`--save_mot_txts`表示保存跟踪结果的txt文件，或`--save_images`表示保存跟踪结果可视化图片。
+
+## 引用
+```
+@article{zhang2020fair,
+  title={FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking},
+  author={Zhang, Yifu and Wang, Chunyu and Wang, Xinggang and Zeng, Wenjun and Liu, Wenyu},
+  journal={arXiv preprint arXiv:2004.01888},
+  year={2020}
+}
+
+@InProceedings{bdd100k,
+    author = {Yu, Fisher and Chen, Haofeng and Wang, Xin and Xian, Wenqi and Chen,
+              Yingying and Liu, Fangchen and Madhavan, Vashisht and Darrell, Trevor},
+    title = {BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning},
+    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+    month = {June},
+    year = {2020}
+}
+
+@INPROCEEDINGS{Geiger2012CVPR,
+  author = {Andreas Geiger and Philip Lenz and Raquel Urtasun},
+  title = {Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},
+  booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year = {2012}
+}
+
+@article{zhu2018vision,
+  title={Vision meets drones: A challenge},
+  author={Zhu, Pengfei and Wen, Longyin and Bian, Xiao and Ling, Haibin and Hu, Qinghua},
+  journal={arXiv preprint arXiv:1804.07437},
+  year={2018}
+}
+@article{zhu2020vision,
+  title={Vision Meets Drones: Past, Present and Future},
+  author={Zhu, Pengfei and Wen, Longyin and Du, Dawei and Bian, Xiao and Hu, Qinghua and Ling, Haibin},
+  journal={arXiv preprint arXiv:2001.06303},
+  year={2020}
+}
+```
--- a/configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml
+++ b/configs/mot/vehicle/fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml
+_BASE_: [
+  '../fairmot/fairmot_dla34_30e_1088x608.yml'
+]
+
+weights: output/fairmot_dla34_30e_1088x608_bdd100kmot_vehicle/model_final
+
+# for MOT training
+TrainDataset:
+  !MOTDataSet
+    dataset_dir: dataset/mot
+    image_lists: ['bdd100kmot_vehicle.train']
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']
+
+# for MOT evaluation
+# If you want to change the MOT evaluation dataset, please modify 'data_root'
+EvalMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    data_root: bdd100kmot_vehicle/images/val
+    keep_ori_im: False # set True if save visualization images or video, or used in DeepSORT
+
+# for MOT video inference
+TestMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    keep_ori_im: True # set True if save visualization images or video
--- a/configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml
+++ b/configs/mot/kitticars/fairmot_dla34_30e_1088x608_kitticars.yml
@@ -3,13 +3,13 @@ _BASE_: [
 ]

 metric: KITTI
-weights: output/fairmot_dla34_30e_1088x608_kitticars/model_final
+weights: output/fairmot_dla34_30e_1088x608_kitti_vehicle/model_final

 # for MOT training
 TrainDataset:
  !MOTDataSet
    dataset_dir: dataset/mot
-    image_lists: ['kitticars.train']
+    image_lists: ['kitti_vehicle.train']
    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']

 # for MOT evaluation
@@ -17,7 +17,7 @@ TrainDataset:
 EvalMOTDataset:
  !MOTImageFolder
    dataset_dir: dataset/mot
-    data_root: kitticars/images/test
+    data_root: kitti_vehicle/images/train
    keep_ori_im: False # set True if save visualization images or video, or used in DeepSORT

 # for MOT video inference

--- a/configs/mot/vehicle/fairmot_dla34_30e_1088x608_visdrone_vehicle.yml
+++ b/configs/mot/vehicle/fairmot_dla34_30e_1088x608_visdrone_vehicle.yml
+_BASE_: [
+  '../fairmot/fairmot_dla34_30e_1088x608.yml'
+]
+
+weights: output/fairmot_dla34_30e_1088x608_visdrone_vehicle/model_final
+
+# for MOT training
+TrainDataset:
+  !MOTDataSet
+    dataset_dir: dataset/mot
+    image_lists: ['visdrone_vehicle.train']
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']
+
+# for MOT evaluation
+# If you want to change the MOT evaluation dataset, please modify 'data_root'
+EvalMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    data_root: visdrone_vehicle/images/val
+    keep_ori_im: False # set True if save visualization images or video, or used in DeepSORT
+
+# for MOT video inference
+TestMOTDataset:
+  !MOTImageFolder
+    dataset_dir: dataset/mot
+    keep_ori_im: True # set True if save visualization images or video
--- a/configs/mot/vehicle/tools/bdd100kmot/bdd100k2mot.py
+++ b/configs/mot/vehicle/tools/bdd100kmot/bdd100k2mot.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import glob
+import os
+import os.path as osp
+import cv2
+import random
+import numpy as np
+import argparse
+
+
+def mkdir_if_missing(d):
+    if not osp.exists(d):
+        os.makedirs(d)
+
+
+def bdd2mot_tracking(img_dir, label_dir, save_img_dir, save_label_dir):
+    label_jsons = os.listdir(label_dir)
+    for label_json in tqdm(label_jsons):
+        with open(os.path.join(label_dir, label_json)) as f:
+            labels_json = json.load(f)
+            for label_json in labels_json:
+                img_name = label_json['name']
+                video_name = label_json['videoName']
+                labels = label_json['labels']
+                txt_string = ""
+                for label in labels:
+                    category = label['category']
+                    x1 = label['box2d']['x1']
+                    x2 = label['box2d']['x2']
+                    y1 = label['box2d']['y1']
+                    y2 = label['box2d']['y2']
+                    width = x2 - x1
+                    height = y2 - y1
+                    x_center = (x1 + x2) / 2. / args.width
+                    y_center = (y1 + y2) / 2. / args.height
+                    width /= args.width
+                    height /= args.height
+                    identity = int(label['id'])
+                    # [class] [identity] [x_center] [y_center] [width] [height]
+                    txt_string += "{} {} {} {} {} {}\n".format(
+                        attr_id_dict[category], identity, x_center, y_center,
+                        width, height)
+
+                fn_label = os.path.join(save_label_dir, img_name[:-4] + '.txt')
+                source_img = os.path.join(img_dir, video_name, img_name)
+                target_img = os.path.join(save_img_dir, img_name)
+                with open(fn_label, 'w') as f:
+                    f.write(txt_string)
+                os.system('cp {} {}'.format(source_img, target_img))
+
+
+def transBBOx(bbox):
+    # bbox --> cx cy w h
+    bbox = list(map(lambda x: float(x), bbox))
+    bbox[0] = (bbox[0] - bbox[2] / 2) * 1280
+    bbox[1] = (bbox[1] - bbox[3] / 2) * 720
+    bbox[2] = bbox[2] * 1280
+    bbox[3] = bbox[3] * 720
+
+    bbox = list(map(lambda x: str(x), bbox))
+    return bbox
+
+
+def genSingleImageMot(inputPath, classes=[]):
+    labelPaths = glob.glob(inputPath + '/*.txt')
+    labelPaths = sorted(labelPaths)
+    allLines = []
+    result = {}
+    for labelPath in labelPaths:
+        frame = str(int(labelPath.split('-')[-1].replace('.txt', '')))
+        with open(labelPath, 'r') as labelPathFile:
+            lines = labelPathFile.readlines()
+            for line in lines:
+                line = line.replace('\n', '')
+                lineArray = line.split(' ')
+                if len(classes) > 0:
+                    if lineArray[0] in classes:
+                        lineArray.append(frame)
+                        allLines.append(lineArray)
+                else:
+                    lineArray.append(frame)
+                    allLines.append(lineArray)
+    resultMap = {}
+    for line in allLines:
+        if line[1] not in resultMap.keys():
+            resultMap[line[1]] = []
+        resultMap[line[1]].append(line)
+    mot_gt = []
+    id_idx = 0
+    for rid in resultMap.keys():
+        id_idx += 1
+        for id_line in resultMap[rid]:
+            mot_line = []
+            mot_line.append(id_line[-1])
+            mot_line.append(str(id_idx))
+            id_line_temp = transBBOx(id_line[2:6])
+            mot_line.extend(id_line_temp)
+            mot_line.append('1')
+            # mot_line.append(id_line[0]) # origin class 
+            mot_line.append('1')  # permanent class  => 1
+            mot_line.append('1')
+            mot_gt.append(mot_line)
+
+    result = list(map(lambda line: str.join(',', line), mot_gt))
+    resultStr = str.join('\n', result)
+    return resultStr
+
+
+def writeGt(inputPath, outPath, classes=[]):
+    singleImageResult = genSingleImageMot(inputPath, classes=classes)
+    outPathFile = outPath + '/gt.txt'
+    mkdir_if_missing(outPath)
+    with open(outPathFile, 'w') as gtFile:
+        gtFile.write(singleImageResult)
+
+
+def genSeqInfo(seqInfoPath):
+    name = seqInfoPath.split('/')[-2]
+    img1Path = osp.join(str.join('/', seqInfoPath.split('/')[0:-1]), 'img1')
+    seqLength = len(glob.glob(img1Path + '/*.jpg'))
+    seqInfoStr = f'''[Sequence]\nname={name}\nimDir=img1\nframeRate=30\nseqLength={seqLength}\nimWidth=1280\nimHeight=720\nimExt=.jpg'''
+    with open(seqInfoPath, 'w') as seqFile:
+        seqFile.write(seqInfoStr)
+
+
+def genMotGt(dataDir, classes=[]):
+    seqLists = sorted(glob.glob(dataDir))
+    for seqList in seqLists:
+        inputPath = osp.join(seqList, 'img1')
+        outputPath = seqList.replace('labels_with_ids', 'images')
+        outputPath = osp.join(outputPath, 'gt')
+        mkdir_if_missing(outputPath)
+        print('processing...', outputPath)
+        writeGt(inputPath, outputPath, classes=classes)
+        seqList = seqList.replace('labels_with_ids', 'images')
+        seqInfoPath = osp.join(seqList, 'seqinfo.ini')
+        genSeqInfo(seqInfoPath)
+
+
+def updateSeqInfo(dataDir, phase):
+    seqPath = osp.join(dataDir, 'labels_with_ids', phase)
+    seqList = glob.glob(seqPath + '/*')
+    for seqName in seqList:
+        print('seqName=>', seqName)
+        seqName_img1_dir = osp.join(seqName, 'img1')
+        txtLength = glob.glob(seqName_img1_dir + '/*.txt')
+        name = seqName.split('/')[-1].replace('.jpg', '').replace('.txt', '')
+        seqLength = len(txtLength)
+        seqInfoStr = f'''[Sequence]\nname={name}\nimDir=img1\nframeRate=30\nseqLength={seqLength}\nimWidth=1280\nimHeight=720\nimExt=.jpg'''
+        seqInfoPath = seqName_img1_dir.replace('labels_with_ids', 'images')
+        seqInfoPath = seqInfoPath.replace('/img1', '')
+        seqInfoPath = seqInfoPath + '/seqinfo.ini'
+        with open(seqInfoPath, 'w') as seqFile:
+            seqFile.write(seqInfoStr)
+
+
+def VisualDataset(datasetPath, phase='train', seqName='', frameId=1):
+    trainPath = osp.join(datasetPath, 'labels_with_ids', phase)
+    seq1Paths = osp.join(trainPath, seqName)
+    seq_img1_path = osp.join(seq1Paths, 'img1')
+    label_with_idPath = osp.join(seq_img1_path, seqName + '-' + '%07d' %
+                                 frameId) + '.txt'
+    image_path = label_with_idPath.replace('labels_with_ids', 'images').replace(
+        '.txt', '.jpg')
+
+    seqInfoPath = str.join('/', image_path.split('/')[:-2])
+    seqInfoPath = seqInfoPath + '/seqinfo.ini'
+    seq_info = open(seqInfoPath).read()
+    width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+        '\nimHeight')])
+    height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+        '\nimExt')])
+
+    with open(label_with_idPath, 'r') as label:
+        allLines = label.readlines()
+        images = cv2.imread(image_path)
+        print('image_path => ', image_path)
+        for line in allLines:
+            line = line.split(' ')
+            line = list(map(lambda x: float(x), line))
+            c1, c2, w, h = line[2:6]
+            x1 = c1 - w / 2
+            x2 = c2 - h / 2
+            x3 = c1 + w / 2
+            x4 = c2 + h / 2
+            cv2.rectangle(
+                images, (int(x1 * width), int(x2 * height)),
+                (int(x3 * width), int(x4 * height)), (255, 0, 0),
+                thickness=2)
+        cv2.imwrite('test.jpg', images)
+
+
+def VisualGt(dataPath, phase='train'):
+    seqList = sorted(glob.glob(osp.join(dataPath, 'images', phase) + '/*'))
+    seqIndex = random.randint(0, len(seqList) - 1)
+    seqPath = seqList[seqIndex]
+    gt_path = osp.join(seqPath, 'gt', 'gt.txt')
+    img_list_path = sorted(glob.glob(osp.join(seqPath, 'img1', '*.jpg')))
+    imgIndex = random.randint(0, len(img_list_path))
+    img_Path = img_list_path[imgIndex]
+    #
+    frame_value = img_Path.split('/')[-1].replace('.jpg', '')
+    frame_value = frame_value.split('-')[-1]
+    frame_value = int(frame_value)
+    seqNameStr = img_Path.split('/')[-1].replace('.jpg', '').replace('img', '')
+    frame_value = int(seqNameStr.split('-')[-1])
+    print('frame_value => ', frame_value)
+    gt_value = np.loadtxt(gt_path, dtype=float, delimiter=',')
+    gt_value = gt_value[gt_value[:, 0] == frame_value]
+
+    get_list = gt_value.tolist()
+    img = cv2.imread(img_Path)
+
+    colors = [[255, 0, 0], [255, 255, 0], [255, 0, 255], [0, 255, 0],
+              [0, 255, 255], [0, 0, 255]]
+    for seq, _id, pl, pt, w, h, _, bbox_class, _ in get_list:
+        pl, pt, w, h = int(pl), int(pt), int(w), int(h)
+        print('pl,pt,w,h => ', pl, pt, w, h)
+        cv2.putText(img,
+                    str(bbox_class), (pl, pt), cv2.FONT_HERSHEY_PLAIN, 2,
+                    colors[int(bbox_class - 1)])
+        cv2.rectangle(
+            img, (pl, pt), (pl + w, pt + h),
+            colors[int(bbox_class - 1)],
+            thickness=2)
+    cv2.imwrite('testGt.jpg', img)
+    print(seqPath, frame_value)
+    return seqPath.split('/')[-1], frame_value
+
+
+def gen_image_list(dataPath, datType):
+    inputPath = f'{dataPath}/labels_with_ids/{datType}'
+    pathList = sorted(glob.glob(inputPath + '/*'))
+    print(pathList)
+    allImageList = []
+    for pathSingle in pathList:
+        imgList = sorted(glob.glob(osp.join(pathSingle, 'img1', '*.txt')))
+        for imgPath in imgList:
+            imgPath = imgPath.replace('labels_with_ids', 'images').replace(
+                '.txt', '.jpg')
+            allImageList.append(imgPath)
+    with open(f'{dataPath}.{datType}', 'w') as image_list_file:
+        allImageListStr = str.join('\n', allImageList)
+        image_list_file.write(allImageListStr)
+
+
+def formatOrigin(datapath, phase):
+    label_with_idPath = osp.join(datapath, 'labels_with_ids', phase)
+    print(label_with_idPath)
+    for txtList in sorted(glob.glob(label_with_idPath + '/*.txt')):
+        print(txtList)
+        seqName = txtList.split('/')[-1]
+        seqName = str.join('-', seqName.split('-')[0:-1]).replace('.txt', '')
+        seqPath = osp.join(label_with_idPath, seqName, 'img1')
+        mkdir_if_missing(seqPath)
+        os.system(f'mv {txtList} {seqPath}')
+
+
+def copyImg(fromRootPath, toRootPath, phase):
+    fromPath = osp.join(fromRootPath, 'images', phase)
+    toPathSeqPath = osp.join(toRootPath, 'labels_with_ids', phase)
+    seqList = sorted(glob.glob(toPathSeqPath + '/*'))
+    for seqPath in seqList:
+        seqName = seqPath.split('/')[-1]
+        imgTxtList = sorted(glob.glob(osp.join(seqPath, 'img1') + '/*.txt'))
+        img_toPathSeqPath = osp.join(seqPath, 'img1')
+        img_toPathSeqPath = img_toPathSeqPath.replace('labels_with_ids',
+                                                      'images')
+        mkdir_if_missing(img_toPathSeqPath)
+
+        for imgTxt in imgTxtList:
+            imgName = imgTxt.split('/')[-1].replace('.txt', '.jpg')
+            imgfromPath = osp.join(fromPath, seqName, imgName)
+            print(f'cp {imgfromPath} {img_toPathSeqPath}')
+            os.system(f'cp {imgfromPath} {img_toPathSeqPath}')
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='BDD100K to MOT format')
+    parser.add_argument("--data_path", default='/path/to/bdd100k')
+    parser.add_argument("--phase", default='train')
+    parser.add_argument("--classes", default='2,3,4,9,10')
+
+    parser.add_argument("--img_dir", default="/path/to/bdd/image/")
+    parser.add_argument("--label_dir", default="/path/to/bdd/label/")
+    parser.add_argument("--save_path", default="/save/path")
+    parser.add_argument("--height", default=720)
+    parser.add_argument("--width", default=1280)
+    args = parser.parse_args()
+
+    ### for bdd tracking dataset
+    attr_dict = dict()
+    attr_dict["categories"] = [{
+        "supercategory": "none",
+        "id": 0,
+        "name": "pedestrian"
+    }, {
+        "supercategory": "none",
+        "id": 1,
+        "name": "rider"
+    }, {
+        "supercategory": "none",
+        "id": 2,
+        "name": "car"
+    }, {
+        "supercategory": "none",
+        "id": 3,
+        "name": "truck"
+    }, {
+        "supercategory": "none",
+        "id": 4,
+        "name": "bus"
+    }, {
+        "supercategory": "none",
+        "id": 5,
+        "name": "train"
+    }, {
+        "supercategory": "none",
+        "id": 6,
+        "name": "motorcycle"
+    }, {
+        "supercategory": "none",
+        "id": 7,
+        "name": "bicycle"
+    }, {
+        "supercategory": "none",
+        "id": 8,
+        "name": "other person"
+    }, {
+        "supercategory": "none",
+        "id": 9,
+        "name": "trailer"
+    }, {
+        "supercategory": "none",
+        "id": 10,
+        "name": "other vehicle"
+    }]
+    attr_id_dict = {i['name']: i['id'] for i in attr_dict['categories']}
+
+    # create BDD training set tracking in MOT format
+    print('Loading and converting training set...')
+    train_img_dir = os.path.join(args.img_dir, 'train')
+    train_label_dir = os.path.join(args.label_dir, 'train')
+    save_img_dir = os.path.join(args.save_path, 'images', 'train')
+    save_label_dir = os.path.join(args.save_path, 'labels_with_ids', 'train')
+    if not os.path.exists(save_img_dir): os.makedirs(save_img_dir)
+    if not os.path.exists(save_label_dir): os.makedirs(save_label_dir)
+    bdd2mot_tracking(train_img_dir, train_label_dir, save_img_dir,
+                     save_label_dir)
+
+    # create BDD validation set tracking in MOT format
+    print('Loading and converting validation set...')
+    val_img_dir = os.path.join(args.img_dir, 'val')
+    val_label_dir = os.path.join(args.label_dir, 'val')
+    save_img_dir = os.path.join(args.save_path, 'images', 'val')
+    save_label_dir = os.path.join(args.save_path, 'labels_with_ids', 'val')
+    if not os.path.exists(save_img_dir): os.makedirs(save_img_dir)
+    if not os.path.exists(save_label_dir): os.makedirs(save_label_dir)
+    bdd2mot_tracking(val_img_dir, val_label_dir, save_img_dir, save_label_dir)
+
+    # gen gt file
+    dataPath = args.data_path
+    phase = args.phase
+    classes = args.classes.split(',')
+    formatOrigin(osp.join(dataPath, 'bdd100kmot_vehicle'), phase)
+    dataDir = osp.join(
+        osp.join(dataPath, 'bdd100kmot_vehicle'), 'labels_with_ids',
+        phase) + '/*'
+    genMotGt(dataDir, classes=classes)
+    copyImg(dataPath, osp.join(dataPath, 'bdd100kmot_vehicle'), phase)
+    updateSeqInfo(osp.join(dataPath, 'bdd100kmot_vehicle'), phase)
+    gen_image_list(osp.join(dataPath, 'bdd100kmot_vehicle'), phase)
+    os.system(f'rm -r {dataPath}/bdd100kmot_vehicle/images/' + phase + '/*.jpg')
--- a/configs/mot/vehicle/tools/bdd100kmot/gen_bdd100kmot_vehicle.sh
+++ b/configs/mot/vehicle/tools/bdd100kmot/gen_bdd100kmot_vehicle.sh
+data_path=bdd100k_path
+img_dir=${data_path}/images
+label_dir=${data_path}/labels
+save_path=${data_path}/bdd100kmot_vehicle
+
+phasetrain=train
+phaseval=val
+classes=2,3,4,9,10
+
+# gen mot dataste
+python bdd100k2mot.py --data_path=${data_path}  --phase=${phasetrain} --classes=${classes} --img_dir=${img_dir} --label_dir=${label_dir} --save_path=${save_path}
+python bdd100k2mot.py --data_path=${data_path} --phase=${phaseval} --classes=${classes} --img_dir=${img_dir} --label_dir=${label_dir} --save_path=${save_path}
+
+# gen new labels_with_ids
+python gen_labels_MOT.py --mot_data=${data_path} --phase=${phasetrain}
+python gen_labels_MOT.py --mot_data=${data_path} --phase=${phaseval}
--- a/configs/mot/vehicle/tools/bdd100kmot/gen_labels_MOT.py
+++ b/configs/mot/vehicle/tools/bdd100kmot/gen_labels_MOT.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import os.path as osp
+import numpy as np
+import argparse
+
+
+def mkdirs(d):
+    if not osp.exists(d):
+        os.makedirs(d)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='BDD100K to MOT format')
+    parser.add_argument(
+        "--mot_data", default='/paddle/dataset/bdd100kmot/bdd100k_small')
+    parser.add_argument("--phase", default='train')
+    args = parser.parse_args()
+
+    MOT_data = args.mot_data
+    phase = args.phase
+    seq_root = osp.join(MOT_data, 'bdd100kmot_vehicle', 'images', phase)
+    label_root = osp.join(MOT_data, 'bdd100kmot_vehicle', 'labels_with_ids',
+                          phase)
+    mkdirs(label_root)
+    seqs = [s for s in os.listdir(seq_root)]
+    tid_curr = 0
+    tid_last = -1
+
+    os.system(f'rm -r {MOT_data}/bdd100kmot_vehicle/labels_with_ids')
+    for seq in seqs:
+        print('seq => ', seq)
+        seq_info = open(osp.join(seq_root, seq, 'seqinfo.ini')).read()
+        seq_width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+            '\nimHeight')])
+        seq_height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+            '\nimExt')])
+
+        gt_txt = osp.join(seq_root, seq, 'gt', 'gt.txt')
+        gt = np.loadtxt(gt_txt, dtype=np.float64, delimiter=',')
+
+        seq_label_root = osp.join(label_root, seq, 'img1')
+        mkdirs(seq_label_root)
+
+        for fid, tid, x, y, w, h, mark, label, _ in gt:
+            fid = int(fid)
+            tid = int(tid)
+            if not tid == tid_last:
+                tid_curr += 1
+                tid_last = tid
+            x += w / 2
+            y += h / 2
+            label_fpath = osp.join(seq_label_root,
+                                   seq + '-' + '{:07d}.txt'.format(fid))
+            label_str = '0 {:d} {:.6f} {:.6f} {:.6f} {:.6f}\n'.format(
+                tid_curr, x / seq_width, y / seq_height, w / seq_width,
+                h / seq_height)
+            with open(label_fpath, 'a') as f:
+                f.write(label_str)
--- a/configs/mot/vehicle/tools/visdrone/visdrone2mot.py
+++ b/configs/mot/vehicle/tools/visdrone/visdrone2mot.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import glob
+import os
+import os.path as osp
+import cv2
+import argparse
+import numpy as np
+import random
+
+# The object category indicates the type of annotated object, 
+# (i.e., ignored regions(0), pedestrian(1), people(2), bicycle(3), car(4), van(5), truck(6), tricycle(7), awning-tricycle(8), bus(9), motor(10),others(11))
+
+
+def mkdir_if_missing(d):
+    if not osp.exists(d):
+        os.makedirs(d)
+
+
+def genGtFile(seqPath, outPath, classes=[]):
+    id_idx = 0
+    old_idx = -1
+    with open(seqPath, 'r') as singleSeqFile:
+        motLine = []
+        allLines = singleSeqFile.readlines()
+        for line in allLines:
+            line = line.replace('\n', '')
+            line = line.split(',')
+            # exclude occlusion!='2'
+            if line[-1] != '2' and line[7] in classes:
+                if old_idx != int(line[1]):
+                    id_idx += 1
+                    old_idx = int(line[1])
+                newLine = line[0:6]
+                newLine[1] = str(id_idx)
+                newLine.append('1')
+                if (len(classes) > 1):
+                    class_index = str(classes.index(line[7]) + 1)
+                    newLine.append(class_index)
+                else:
+                    newLine.append('1')  # use permenant class '1'
+                newLine.append('1')
+                motLine.append(newLine)
+        mkdir_if_missing(outPath)
+        gtFilePath = osp.join(outPath, 'gt.txt')
+        with open(gtFilePath, 'w') as gtFile:
+            motLine = list(map(lambda x: str.join(',', x), motLine))
+            motLineStr = str.join('\n', motLine)
+            gtFile.write(motLineStr)
+
+
+def genSeqInfo(img1Path, seqName):
+    imgPaths = glob.glob(img1Path + '/*.jpg')
+    seqLength = len(imgPaths)
+    if seqLength > 0:
+        image1 = cv2.imread(imgPaths[0])
+        imgHeight = image1.shape[0]
+        imgWidth = image1.shape[1]
+    else:
+        imgHeight = 0
+        imgWidth = 0
+    seqInfoStr = f'''[Sequence]\nname={seqName}\nimDir=img1\nframeRate=30\nseqLength={seqLength}\nimWidth={imgWidth}\nimHeight={imgHeight}\nimExt=.jpg'''
+    seqInfoPath = img1Path.replace('/img1', '')
+    with open(seqInfoPath + '/seqinfo.ini', 'w') as seqFile:
+        seqFile.write(seqInfoStr)
+
+
+def copyImg(img1Path, gtTxtPath, outputFileName):
+    with open(gtTxtPath, 'r') as gtFile:
+        allLines = gtFile.readlines()
+        imgList = []
+        for line in allLines:
+            imgIdx = int(line.split(',')[0])
+            if imgIdx not in imgList:
+                imgList.append(imgIdx)
+                seqName = gtTxtPath.replace('./{}/'.format(outputFileName),
+                                            '').replace('/gt/gt.txt', '')
+                sourceImgPath = osp.join('./sequences', seqName,
+                                         '{:07d}.jpg'.format(imgIdx))
+                os.system(f'cp {sourceImgPath} {img1Path}')
+
+
+def genMotLabels(datasetPath, outputFileName, classes=['2']):
+    mkdir_if_missing(osp.join(datasetPath, outputFileName))
+    annotationsPath = osp.join(datasetPath, 'annotations')
+    annotationsList = glob.glob(osp.join(annotationsPath, '*.txt'))
+    for annotationPath in annotationsList:
+        seqName = annotationPath.split('/')[-1].replace('.txt', '')
+        mkdir_if_missing(osp.join(datasetPath, outputFileName, seqName, 'gt'))
+        mkdir_if_missing(osp.join(datasetPath, outputFileName, seqName, 'img1'))
+        genGtFile(annotationPath,
+                  osp.join(datasetPath, outputFileName, seqName, 'gt'), classes)
+        img1Path = osp.join(datasetPath, outputFileName, seqName, 'img1')
+        gtTxtPath = osp.join(datasetPath, outputFileName, seqName, 'gt/gt.txt')
+        copyImg(img1Path, gtTxtPath, outputFileName)
+        genSeqInfo(img1Path, seqName)
+
+
+def deleteFileWhichImg1IsEmpty(mot16Path, dataType='train'):
+    path = mot16Path
+    data_images_train = osp.join(path, 'images', f'{dataType}')
+    data_images_train_seqs = glob.glob(data_images_train + '/*')
+    if (len(data_images_train_seqs) == 0):
+        print('dataset is empty!')
+    for data_images_train_seq in data_images_train_seqs:
+        data_images_train_seq_img1 = osp.join(data_images_train_seq, 'img1')
+        if len(glob.glob(data_images_train_seq_img1 + '/*.jpg')) == 0:
+            print(f"os.system(rm -rf {data_images_train_seq})")
+            os.system(f'rm -rf {data_images_train_seq}')
+
+
+def formatMot16Path(dataPath, pathType='train'):
+    train_path = osp.join(dataPath, 'images', pathType)
+    mkdir_if_missing(train_path)
+    os.system(f'mv {dataPath}/* {train_path}')
+
+
+def VisualGt(dataPath, phase='train'):
+    seqList = sorted(glob.glob(osp.join(dataPath, 'images', phase) + '/*'))
+    seqIndex = random.randint(0, len(seqList) - 1)
+    seqPath = seqList[seqIndex]
+    gt_path = osp.join(seqPath, 'gt', 'gt.txt')
+    img_list_path = sorted(glob.glob(osp.join(seqPath, 'img1', '*.jpg')))
+    imgIndex = random.randint(0, len(img_list_path))
+    img_Path = img_list_path[imgIndex]
+    frame_value = int(img_Path.split('/')[-1].replace('.jpg', ''))
+    gt_value = np.loadtxt(gt_path, dtype=int, delimiter=',')
+    gt_value = gt_value[gt_value[:, 0] == frame_value]
+    get_list = gt_value.tolist()
+    img = cv2.imread(img_Path)
+    colors = [[255, 0, 0], [255, 255, 0], [255, 0, 255], [0, 255, 0],
+              [0, 255, 255], [0, 0, 255]]
+    for seq, _id, pl, pt, w, h, _, bbox_class, _ in get_list:
+        cv2.putText(img,
+                    str(bbox_class), (pl, pt), cv2.FONT_HERSHEY_PLAIN, 2,
+                    colors[bbox_class - 1])
+        cv2.rectangle(
+            img, (pl, pt), (pl + w, pt + h),
+            colors[bbox_class - 1],
+            thickness=2)
+    cv2.imwrite('testGt.jpg', img)
+
+
+def VisualDataset(datasetPath, phase='train', seqName='', frameId=1):
+    trainPath = osp.join(datasetPath, 'labels_with_ids', phase)
+    seq1Paths = osp.join(trainPath, seqName)
+    seq_img1_path = osp.join(seq1Paths, 'img1')
+    label_with_idPath = osp.join(seq_img1_path, '%07d' % frameId) + '.txt'
+    image_path = label_with_idPath.replace('labels_with_ids', 'images').replace(
+        '.txt', '.jpg')
+    seqInfoPath = str.join('/', image_path.split('/')[:-2])
+    seqInfoPath = seqInfoPath + '/seqinfo.ini'
+    seq_info = open(seqInfoPath).read()
+    width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+        '\nimHeight')])
+    height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+        '\nimExt')])
+
+    with open(label_with_idPath, 'r') as label:
+        allLines = label.readlines()
+        images = cv2.imread(image_path)
+        for line in allLines:
+            line = line.split(' ')
+            line = list(map(lambda x: float(x), line))
+            c1, c2, w, h = line[2:6]
+            x1 = c1 - w / 2
+            x2 = c2 - h / 2
+            x3 = c1 + w / 2
+            x4 = c2 + h / 2
+            cv2.rectangle(
+                images, (int(x1 * width), int(x2 * height)),
+                (int(x3 * width), int(x4 * height)), (255, 0, 0),
+                thickness=2)
+        cv2.imwrite('test.jpg', images)
+
+
+def gen_image_list(dataPath, datType):
+    inputPath = f'{dataPath}/images/{datType}'
+    pathList = glob.glob(inputPath + '/*')
+    pathList = sorted(pathList)
+    allImageList = []
+    for pathSingle in pathList:
+        imgList = sorted(glob.glob(osp.join(pathSingle, 'img1', '*.jpg')))
+        for imgPath in imgList:
+            allImageList.append(imgPath)
+    with open(f'{dataPath}.{datType}', 'w') as image_list_file:
+        allImageListStr = str.join('\n', allImageList)
+        image_list_file.write(allImageListStr)
+
+
+def gen_labels_mot(MOT_data, phase='train'):
+    seq_root = './{}/images/{}'.format(MOT_data, phase)
+    label_root = './{}/labels_with_ids/{}'.format(MOT_data, phase)
+    mkdir_if_missing(label_root)
+    seqs = [s for s in os.listdir(seq_root)]
+    print('seqs => ', seqs)
+    tid_curr = 0
+    tid_last = -1
+    for seq in seqs:
+        seq_info = open(osp.join(seq_root, seq, 'seqinfo.ini')).read()
+        seq_width = int(seq_info[seq_info.find('imWidth=') + 8:seq_info.find(
+            '\nimHeight')])
+        seq_height = int(seq_info[seq_info.find('imHeight=') + 9:seq_info.find(
+            '\nimExt')])
+
+        gt_txt = osp.join(seq_root, seq, 'gt', 'gt.txt')
+        gt = np.loadtxt(gt_txt, dtype=np.float64, delimiter=',')
+
+        seq_label_root = osp.join(label_root, seq, 'img1')
+        mkdir_if_missing(seq_label_root)
+
+        for fid, tid, x, y, w, h, mark, label, _ in gt:
+            # if mark == 0 or not label == 1: 
+            #     continue
+            fid = int(fid)
+            tid = int(tid)
+            if not tid == tid_last:
+                tid_curr += 1
+                tid_last = tid
+            x += w / 2
+            y += h / 2
+            label_fpath = osp.join(seq_label_root, '{:07d}.txt'.format(fid))
+            label_str = '0 {:d} {:.6f} {:.6f} {:.6f} {:.6f}\n'.format(
+                tid_curr, x / seq_width, y / seq_height, w / seq_width,
+                h / seq_height)
+            with open(label_fpath, 'a') as f:
+                f.write(label_str)
+
+
+def parse_arguments():
+    parser = argparse.ArgumentParser(description='input method')
+    parser.add_argument("--transMot", type=bool, default=False)
+    parser.add_argument("--genMot", type=bool, default=False)
+    parser.add_argument("--formatMotPath", type=bool, default=False)
+    parser.add_argument("--deleteEmpty", type=bool, default=False)
+    parser.add_argument("--genLabelsMot", type=bool, default=False)
+    parser.add_argument("--genImageList", type=bool, default=False)
+    parser.add_argument("--visualImg", type=bool, default=False)
+    parser.add_argument("--visualGt", type=bool, default=False)
+    parser.add_argument("--data_name", type=str, default='visdrone_vehicle')
+    parser.add_argument("--phase", type=str, default='train')
+    parser.add_argument("--classes", type=str, default='4,5,6,9')
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_arguments()
+    classes = args.classes.split(',')
+    datasetPath = './'
+    dataName = args.data_name
+    phase = args.phase
+    if args.transMot:
+        genMotLabels(datasetPath, dataName, classes)
+        formatMot16Path(dataName, pathType=phase)
+        mot16Path = f'./{dataName}'
+        deleteFileWhichImg1IsEmpty(mot16Path, dataType=phase)
+        gen_labels_mot(dataName, phase=phase)
+        gen_image_list(dataName, phase)
+    if args.genMot:
+        genMotLabels(datasetPath, dataName, classes)
+    if args.formatMotPath:
+        formatMot16Path(dataName, pathType=phase)
+    if args.deleteEmpty:
+        mot16Path = f'./{dataName}'
+        deleteFileWhichImg1IsEmpty(mot16Path, dataType=phase)
+    if args.genLabelsMot:
+        gen_labels_mot(dataName, phase=phase)
+    if args.genImageList:
+        gen_image_list(dataName, phase)
+    if args.visualGt:
+        VisualGt(f'./{dataName}', phase)
+    if args.visualImg:
+        seqName = 'uav0000137_00458_v'
+        frameId = 43
+        VisualDataset(
+            f'./{dataName}', phase=phase, seqName=seqName, frameId=frameId)
--- a/ppdet/metrics/mot_metrics.py
+++ b/ppdet/metrics/mot_metrics.py
@@ -37,7 +37,7 @@ __all__ = ['MOTEvaluator', 'MOTMetric', 'JDEDetMetric', 'KITTIMOTMetric']

 def read_mot_results(filename, is_gt=False, is_ignore=False):
    valid_labels = {1}
-    ignore_labels = {2, 7, 8, 12}
+    ignore_labels = {2, 7, 8, 12}  # only in motchallenge datasets like 'MOT16'
    results_dict = dict()
    if os.path.isfile(filename):
        with open(filename, 'r') as f:
@@ -53,11 +53,10 @@ def read_mot_results(filename, is_gt=False, is_ignore=False):
                box_size = float(linelist[4]) * float(linelist[5])

                if is_gt:
-                    if 'MOT16-' in filename or 'MOT17-' in filename or 'MOT15-' in filename or 'MOT20-' in filename:
-                        label = int(float(linelist[7]))
-                        mark = int(float(linelist[6]))
-                        if mark == 0 or label not in valid_labels:
-                            continue
+                    label = int(float(linelist[7]))
+                    mark = int(float(linelist[6]))
+                    if mark == 0 or label not in valid_labels:
+                        continue
                    score = 1
                elif is_ignore:
                    if 'MOT16-' in filename or 'MOT17-' in filename or 'MOT15-' in filename or 'MOT20-' in filename: