[MOT] add hrnet_w18 fairmot (#3857)

* add hrnet18 fairmot * add lite hrnet_w18 * add hrnet dlafpn * update hrnet18_dlafpn modelzoo * hrnet18dlafpn modelzoo * add comments * set neck not none, update modelzoo

[MOT] add hrnet_w18 fairmot (#3857)
* add hrnet18 fairmot * add lite hrnet_w18 * add hrnet dlafpn * update hrnet18_dlafpn modelzoo * hrnet18dlafpn modelzoo * add comments * set neck not none, update modelzoo
afa7c14d · Feng Ni · GitHub · 0eb6f21d · afa7c14d · afa7c14d
14 changed file
--- a/configs/mot/README.md
+++ b/configs/mot/README.md
@@ -118,9 +118,9 @@ If you use a stronger detection model, you can get better results. Each txt is t
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
 | DLA-34(paper)  | 1088x608 |  83.3  |  81.9  |   544  |  3822  |  14095  |     -   |    -   |   -    |
-| DLA-34         | 1088x608 |  83.2  |  83.1  |   499  |  3861  |  14223  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  83.2  |  83.1  |   499  |  3861  |  14223  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](./fairmot/fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](./fairmot/fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](./fairmot/fairmot_dla34_30e_576x320.yml) |


 ### FairMOT Results on MOT-16 Test Set
@@ -128,14 +128,29 @@ If you use a stronger detection model, you can get better results. Each txt is t
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
 | DLA-34(paper)  | 1088x608 |  74.9  |  72.8  |  1074  |    -   |    -   |   25.9   |    -   |   -    |
-| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](./fairmot/fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](./fairmot/fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](./fairmot/fairmot_dla34_30e_576x320.yml) |

 **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 
+### Results on MOT-16 Test Set
+| backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  71.7  |  66.6  |  1340  |  8642  | 41592 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [config](./fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+### Results on MOT-17 Test Set
+| backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  70.7  |  65.7  |  4281  |  22485  | 138468 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [config](./fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+**Notes:**
+ FairMOT HRNetV2-W18 used 8 GPUs for training and mini-batch size as 6 on each GPU, and trained for 30 epoches. Only ImageNet pre-train model is used, and the optimizer adopts Momentum. The crowdhuman dataset is added to the train-set during training.
+
+
 ## Feature Tracking Model

 ### [Head Tracking](./headtracking21/README.md)

--- a/configs/mot/README_cn.md
+++ b/configs/mot/README_cn.md
@@ -119,23 +119,38 @@ wget https://dataset.bj.bcebos.com/mot/det_results_dir.zip
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  下载链接 | 配置文件 |
 | :--------------| :------- | :----: | :----: | :---: | :----: | :---: | :------: | :----: |:----: |
 | DLA-34(paper)  | 1088x608 |  83.3  |  81.9  |  544  |  3822  | 14095 |    -     |   -   |   -   |
-| DLA-34         | 1088x608 |  83.2  |  83.1  |  499  |  3861  | 14223 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  83.2  |  83.1  |  499  |  3861  | 14223 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_576x320.yml) |

 ### FairMOT在MOT-16 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(paper)  | 1088x608 |  74.9  |  72.8  |  1074  |    -   |    -   |   25.9   |    -   |   -    |
-| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](./fairmot/fairmot_dla34_30e_576x320.yml) |

 **注意:**
 FairMOT DLA-34均使用2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。


+### FairMOT轻量级模型
+### 在MOT-16 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 |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  71.7  |  66.6  |  1340  |  8642  | 41592 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [配置文件](./fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+### 在MOT-17 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 |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  70.7  |  65.7  |  4281  |  22485  | 138468 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [配置文件](./fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+**注意:**
+ FairMOT HRNetV2-W18均使用8个GPU进行训练，每个GPU上batch size为4，训练30个epoch，使用的ImageNet预训练，优化器策略采用的是Momentum，并且训练集中加入了crowdhuman数据集一起参与训练。
+
+
 ## 特色垂类跟踪模型

 ### [人头跟踪（Head Tracking)](./headtracking21/README.md)

--- a/configs/mot/fairmot/README.md
+++ b/configs/mot/fairmot/README.md
@@ -19,9 +19,9 @@ English | [简体中文](README_cn.md)
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
 | DLA-34(paper)  | 1088x608 |  83.3  |  81.9  |   544  |  3822  |  14095  |     -   |    -   |   -    |
-| DLA-34         | 1088x608 |  83.2  |  83.1  |   499  |  3861  |  14223  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  83.2  |  83.1  |   499  |  3861  |  14223  |     -   | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](./fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](./fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](./fairmot_dla34_30e_576x320.yml) |


 ### FairMOT Results on MOT-16 Test Set
@@ -29,13 +29,28 @@ English | [简体中文](README_cn.md)
 | backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
 | :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
 | DLA-34(paper)  | 1088x608 |  74.9  |  72.8  |  1074  |    -   |    -   |   25.9   |    -   |   -    |
-| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     | [model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [config](./fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [config](./fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [config](./fairmot_dla34_30e_576x320.yml) |

 **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 
+### Results on MOT-16 Test Set
+| backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  71.7  |  66.6  |  1340  |  8642  | 41592 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [config](./fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+### Results on MOT-17 Test Set
+| backbone       | input shape | MOTA | IDF1 |  IDS  |    FP   |   FN   |    FPS    | download | config |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  70.7  |  65.7  |  4281  |  22485  | 138468 |    -     |[model](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [config](./fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+**Notes:**
+ FairMOT HRNetV2-W18 used 8 GPUs for training and mini-batch size as 6 on each GPU, and trained for 30 epoches. Only ImageNet pre-train model is used, and the optimizer adopts Momentum. The crowdhuman dataset is added to the train-set during training.
+
+
 ## Getting Start

 ### 1. Training

--- a/configs/mot/fairmot/README_cn.md
+++ b/configs/mot/fairmot/README_cn.md
@@ -19,22 +19,38 @@
 |    骨干网络      |  输入尺寸 |  MOTA  |  IDF1  |  IDS  |   FP  |   FN   |   FPS   |  下载链接 | 配置文件 |
 | :--------------| :------- | :----: | :----: | :---: | :----: | :---: | :------: | :----: |:----: |
 | DLA-34(paper)  | 1088x608 |  83.3  |  81.9  |  544  |  3822  | 14095 |    -     |   -   |   -   |
-| DLA-34         | 1088x608 |  83.2  |  83.1  |  499  |  3861  | 14223 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  83.2  |  83.1  |  499  |  3861  | 14223 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  80.8  |  81.1  |  561  |  3643  | 16967 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](./fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  74.0  |  76.1  |  640  |  4989  | 23034 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](./fairmot_dla34_30e_576x320.yml) |

 ### FairMOT在MOT-16 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(paper)  | 1088x608 |  74.9  |  72.8  |  1074  |    -   |    -   |   25.9   |    -   |   -    |
-| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_1088x608.yml) |
-| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_864x480.yml) |
-| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/mot/fairmot/fairmot_dla34_30e_576x320.yml) |
+| DLA-34         | 1088x608 |  75.0  |  74.7  |  919   |  7934  |  36747 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_1088x608.pdparams) | [配置文件](./fairmot_dla34_30e_1088x608.yml) |
+| DLA-34         | 864x480 |  73.0  |  72.6  |  977   |  7578  |  40601 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_864x480.pdparams) | [配置文件](./fairmot_dla34_30e_864x480.yml) |
+| DLA-34         | 576x320 |  69.9  |  70.2  |  1044   |  8869  |  44898 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_dla34_30e_576x320.pdparams) | [配置文件](./fairmot_dla34_30e_576x320.yml) |

 **注意:**
 FairMOT DLA-34均使用2个GPU进行训练，每个GPU上batch size为6，训练30个epoch。

+
+### FairMOT轻量级模型
+### 在MOT-16 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 |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  71.7  |  66.6  |  1340  |  8642  | 41592 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [配置文件](./fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+### 在MOT-17 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 |
+| :--------------| :------- | :----: | :----: | :----: | :----: | :----: | :------: | :----: |:-----: |
+| HRNetV2-W18   | 1088x608 |  70.7  |  65.7  |  4281  |  22485  | 138468 |    -     |[下载链接](https://paddledet.bj.bcebos.com/models/mot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.pdparams) | [配置文件](./fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml) |
+
+**注意:**
+ FairMOT HRNetV2-W18均使用8个GPU进行训练，每个GPU上batch size为4，训练30个epoch，使用的ImageNet预训练，优化器策略采用的是Momentum，并且训练集中加入了crowdhuman数据集一起参与训练。
+
+
 ## 快速开始

 ### 1. 训练

--- a/configs/mot/fairmot/_base_/fairmot_hrnetv2_w18_dlafpn.yml
+++ b/configs/mot/fairmot/_base_/fairmot_hrnetv2_w18_dlafpn.yml
+architecture: FairMOT
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/HRNet_W18_C_pretrained.pdparams
+
+FairMOT:
+  detector: CenterNet
+  reid: FairMOTEmbeddingHead
+  loss: FairMOTLoss
+  tracker: JDETracker
+
+CenterNet:
+  backbone: HRNet
+  head: CenterNetHead
+  post_process: CenterNetPostProcess
+  neck: CenterNetDLAFPN
+  for_mot: True
+
+HRNet:
+  width: 18
+  freeze_at: 0
+  return_idx: [0, 1, 2, 3]
+  upsample: False
+
+CenterNetDLAFPN:
+  down_ratio: 4
+  last_level: 3
+  out_channel: 0
+  first_level: 0
+  dcn_v2: False
+
+CenterNetPostProcess:
+  for_mot: True
+
+JDETracker:
+  conf_thres: 0.4
+  tracked_thresh: 0.4
+  metric_type: cosine
--- a/configs/mot/fairmot/_base_/optimizer_30e_momentum.yml
+++ b/configs/mot/fairmot/_base_/optimizer_30e_momentum.yml
+epoch: 30
+
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [15, 22]
+    use_warmup: True
+  - !BurninWarmup
+    steps: 1000
+
+OptimizerBuilder:
+  optimizer:
+    momentum: 0.9
+    type: Momentum
+  regularizer:
+    factor: 0.0001
+    type: L2
--- a/configs/mot/fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml
+++ b/configs/mot/fairmot/fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml
+_BASE_: [
+  '../../datasets/mot.yml',
+  '../../runtime.yml',
+  '_base_/optimizer_30e_momentum.yml',
+  '_base_/fairmot_hrnetv2_w18_dlafpn.yml',
+  '_base_/fairmot_reader_1088x608.yml',
+]
+
+# for MOT training
+TrainDataset:
+  !MOTDataSet
+    dataset_dir: dataset/mot
+    image_lists: ['mot17.train', 'caltech.all', 'cuhksysu.train', 'prw.train', 'citypersons.train', 'eth.train', 'crowdhuman.train', 'crowdhuman.val']
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_ide']
+
+worker_num: 4
+TrainReader:
+  inputs_def:
+    image_shape: [3, 608, 1088]
+  sample_transforms:
+    - Decode: {}
+    - RGBReverse: {}
+    - AugmentHSV: {}
+    - LetterBoxResize: {target_size: [608, 1088]}
+    - MOTRandomAffine: {reject_outside: False}
+    - RandomFlip: {}
+    - BboxXYXY2XYWH: {}
+    - NormalizeBox: {}
+    - NormalizeImage: {mean: [0, 0, 0], std: [1, 1, 1]}
+    - RGBReverse: {}
+    - Permute: {}
+  batch_transforms:
+    - Gt2FairMOTTarget: {}
+  batch_size: 4
+  shuffle: True
+  drop_last: True
+  use_shared_memory: True
+
+weights: output/fairmot_hrnetv2_w18_dlafpn_30e_1088x608/model_final
--- a/deploy/python/mot_jde_infer.py
+++ b/deploy/python/mot_jde_infer.py
@@ -267,8 +267,8 @@ def predict_video(detector, camera_id):

    if FLAGS.save_images:
        save_dir = os.path.join(FLAGS.output_dir, video_name.split('.')[-2])
-        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(
-            save_dir, out_path)
+        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(save_dir,
+                                                              out_path)
        os.system(cmd_str)
        print('Save video in {}.'.format(out_path))
    else:

--- a/deploy/python/mot_keypoint_unite_infer.py
+++ b/deploy/python/mot_keypoint_unite_infer.py
@@ -216,8 +216,8 @@ def mot_keypoint_unite_predict_video(mot_model,

    if FLAGS.save_images:
        save_dir = os.path.join(FLAGS.output_dir, video_name.split('.')[-2])
-        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(
-            save_dir, out_path)
+        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(save_dir,
+                                                              out_path)
        os.system(cmd_str)
        print('Save video in {}.'.format(out_path))
    else:

--- a/deploy/python/mot_sde_infer.py
+++ b/deploy/python/mot_sde_infer.py
@@ -408,8 +408,8 @@ def predict_video(detector, reid_model, camera_id):

    if FLAGS.save_images:
        save_dir = os.path.join(FLAGS.output_dir, video_name.split('.')[-2])
-        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(
-            save_dir, out_path)
+        cmd_str = 'ffmpeg -f image2 -i {}/%05d.jpg {}'.format(save_dir,
+                                                              out_path)
        os.system(cmd_str)
        print('Save video in {}.'.format(out_path))
    else:

--- a/ppdet/modeling/architectures/centernet.py
+++ b/ppdet/modeling/architectures/centernet.py
@@ -29,7 +29,7 @@ class CenterNet(BaseArch):

    Args:
        backbone (object): backbone instance
-        neck (object): 'CenterDLAFPN' instance
+        neck (object): FPN instance, default None, use 'CenterDLAFPN' in FairMOT
        head (object): 'CenterHead' instance
        post_process (object): 'CenterNetPostProcess' instance
        for_mot (bool): whether return other features used in tracking model
@@ -39,7 +39,7 @@ class CenterNet(BaseArch):
    __inject__ = ['post_process']

    def __init__(self,
-                 backbone='DLA',
+                 backbone,
                 neck='CenterDLAFPN',
                 head='CenterHead',
                 post_process='CenterNetPostProcess',
@@ -56,16 +56,18 @@ class CenterNet(BaseArch):
        backbone = create(cfg['backbone'])

        kwargs = {'input_shape': backbone.out_shape}
-        neck = create(cfg['neck'], **kwargs)
+        neck = cfg['neck'] and create(cfg['neck'], **kwargs)

-        kwargs = {'input_shape': neck.out_shape}
+        out_shape = neck and neck.out_shape or backbone.out_shape
+        kwargs = {'input_shape': out_shape}
        head = create(cfg['head'], **kwargs)

        return {'backbone': backbone, 'neck': neck, "head": head}

    def _forward(self):
-        body_feats = self.backbone(self.inputs)
-        neck_feat = self.neck(body_feats)
+        neck_feat = self.backbone(self.inputs)
+        if self.neck is not None:
+            neck_feat = self.neck(neck_feat)
        head_out = self.head(neck_feat, self.inputs)
        if self.for_mot:
            head_out.update({'neck_feat': neck_feat})

--- a/ppdet/modeling/architectures/fairmot.py
+++ b/ppdet/modeling/architectures/fairmot.py
@@ -53,8 +53,9 @@ class FairMOT(BaseArch):
    @classmethod
    def from_config(cls, cfg, *args, **kwargs):
        detector = create(cfg['detector'])
+        detector_out_shape = detector.neck and detector.neck.out_shape or detector.backbone.out_shape

-        kwargs = {'input_shape': detector.neck.out_shape}
+        kwargs = {'input_shape': detector_out_shape}
        reid = create(cfg['reid'], **kwargs)
        loss = create(cfg['loss'])
        tracker = create(cfg['tracker'])

--- a/ppdet/modeling/backbones/hrnet.py
+++ b/ppdet/modeling/backbones/hrnet.py
@@ -569,6 +569,7 @@ class HRNet(nn.Layer):
        freeze_norm (bool): whether to freeze norm in HRNet
        norm_decay (float): weight decay for normalization layer weights
        return_idx (List): the stage to return
+        upsample (bool): whether to upsample and concat the backbone feats
    """

    def __init__(self,
@@ -577,7 +578,8 @@ class HRNet(nn.Layer):
                 freeze_at=0,
                 freeze_norm=True,
                 norm_decay=0.,
-                 return_idx=[0, 1, 2, 3]):
+                 return_idx=[0, 1, 2, 3],
+                 upsample=False):
        super(HRNet, self).__init__()

        self.width = width
@@ -588,6 +590,7 @@ class HRNet(nn.Layer):
        assert len(return_idx) > 0, "need one or more return index"
        self.freeze_at = freeze_at
        self.return_idx = return_idx
+        self.upsample = upsample

        self.channels = {
            18: [[18, 36], [18, 36, 72], [18, 36, 72, 144]],
@@ -602,8 +605,8 @@ class HRNet(nn.Layer):

        channels_2, channels_3, channels_4 = self.channels[width]
        num_modules_2, num_modules_3, num_modules_4 = 1, 4, 3
-        self._out_channels = channels_4
-        self._out_strides = [4, 8, 16, 32]
+        self._out_channels = [sum(channels_4)] if self.upsample else channels_4
+        self._out_strides = [4] if self.upsample else [4, 8, 16, 32]

        self.conv_layer1_1 = ConvNormLayer(
            ch_in=3,
@@ -695,6 +698,15 @@ class HRNet(nn.Layer):

        st4 = self.st4(tr3)

+        if self.upsample:
+            # Upsampling
+            x0_h, x0_w = st4[0].shape[2:4]
+            x1 = F.upsample(st4[1], size=(x0_h, x0_w), mode='bilinear')
+            x2 = F.upsample(st4[2], size=(x0_h, x0_w), mode='bilinear')
+            x3 = F.upsample(st4[3], size=(x0_h, x0_w), mode='bilinear')
+            x = paddle.concat([st4[0], x1, x2, x3], 1)
+            return x
+
        res = []
        for i, layer in enumerate(st4):
            if i == self.freeze_at:
@@ -706,6 +718,8 @@ class HRNet(nn.Layer):

    @property
    def out_shape(self):
+        if self.upsample:
+            self.return_idx = [0]
        return [
            ShapeSpec(
                channels=self._out_channels[i], stride=self._out_strides[i])

--- a/ppdet/modeling/necks/centernet_fpn.py
+++ b/ppdet/modeling/necks/centernet_fpn.py
@@ -134,8 +134,9 @@ class CenterNetDLAFPN(nn.Layer):
        last_level (int): the last level of input feature fed into the upsamplng block
        out_channel (int): the channel of the output feature, 0 by default means
            the channel of the input feature whose down ratio is `down_ratio`
+        first_level (int): the first level of input feature fed into the upsamplng
+            block, -1 by default and it will be calculated by down_ratio
        dcn_v2 (bool): whether use the DCNv2, true by default
-        
    """

    def __init__(self,
@@ -143,9 +144,11 @@ class CenterNetDLAFPN(nn.Layer):
                 down_ratio=4,
                 last_level=5,
                 out_channel=0,
+                 first_level=-1,
                 dcn_v2=True):
        super(CenterNetDLAFPN, self).__init__()
-        self.first_level = int(np.log2(down_ratio))
+        self.first_level = int(np.log2(
+            down_ratio)) if first_level == -1 else first_level
        self.down_ratio = down_ratio
        self.last_level = last_level
        scales = [2**i for i in range(len(in_channels[self.first_level:]))]