Merge branch 'develop' into pphuman_stgcn

Author: zoooo0820

README.md

@@ -2,17 +2,14 @@
 
 # PaddleVideo
 
-## 近期活动
+## 近期更新
 
-🌟  **1月17号-21号《产业级视频技术与应用案例》** 🌟
-- 【1月17号20:15-21:30】视频技术导论及医疗行业典型案例
-- 【1月18号20:15-21:30】视频内容智能分析和生产解决方案
-- 【1月19号20:15-21:30】体育+安全防范行业中的行为识别
-- 【1月20号20:15-21:30】顶会冠军视频分割算法深度解密
-- 【1月21号20:15-21:30】多模态学习和检索方法
+- 新增骨骼点行为识别模型[CTR-GCN](./docs/zh-CN/model_zoo/recognition/ctrgcn.md).
+- 新增轻量化行为识别模型[MoViNet](./docs/zh-CN/model_zoo/recognition/movinet.md).
+- 新增视频时序分割模型[MS-TCN](./docs/zh-CN/model_zoo/segmentation/mstcn.md)、[ASRF](./docs/zh-CN/model_zoo/segmentation/asrf.md).
 
-👀 **报名链接**: https://paddleqiyeban.wjx.cn/vj/QIValIZ.aspx?udsid=419689
-**课程回放链接**:  https://aistudio.baidu.com/aistudio/course/introduce/6742
+
+👀 🌟  **《产业级视频技术与应用案例》系列课程回放链接**:  https://aistudio.baidu.com/aistudio/course/introduce/6742 🌟
 
 ​																	  💖 **欢迎大家扫码入群讨论** 💖
 <div align="center">
@@ -53,15 +50,15 @@ PaddleVideo是[飞桨官方](https://www.paddlepaddle.org.cn/?fr=paddleEdu_githu
     <td><a href="./docs/zh-CN/model_zoo/recognition/timesformer.md">TimeSformer</a> (Transformer‘)</td>
     <td><a href="./docs/zh-CN/model_zoo/recognition/videoswin.md">VideoSwin</a> (Transformer’)</td>
     <td><a href="./docs/zh-CN/model_zoo/recognition/attention_lstm.md">AttentionLSTM</a> (RNN‘)</td>
-    <td></td>
+    <td><a href="./docs/zh-CN/model_zoo/recognition/movinet.md">MoViNet</a> (Lite‘)</td>
   </tr>
   <tr>
     <td colspan="5" style="font-weight:bold;">基于骨骼点的动作识别方法</td>
   </tr>
   <tr>
-    <td><a href="./docs/zh-CN/model_zoo/recognition/stgcn.md">ST-GCN</a> (Custom’)</td>
-    <td><a href="./docs/zh-CN/model_zoo/recognition/agcn.md">AGCN</a> (Adaptive‘)</td>
-    <td></td>
+    <td><a href="./docs/zh-CN/model_zoo/recognition/stgcn.md">ST-GCN</a> (GCN’)</td>
+    <td><a href="./docs/zh-CN/model_zoo/recognition/agcn.md">AGCN</a> (GCN‘)</td>
+    <td><a href="./docs/zh-CN/model_zoo/recognition/ctrgcn.md">CTR-GCN</a> (GCN‘)</td>
     <td></td>
     <td></td>
   </tr>
@@ -75,6 +72,16 @@ PaddleVideo是[飞桨官方](https://www.paddlepaddle.org.cn/?fr=paddleEdu_githu
     <td></td>
     <td></td>
   </tr>
+  <tr>
+    <td colspan="5" style="font-weight:bold;">视频时序分割</td>
+  </tr>
+  <tr>
+    <td><a href="./docs/zh-CN/model_zoo/segmentation/mstcn.md">MS-TCN</a> </td>
+    <td><a href="./docs/zh-CN/model_zoo/segmentation/asrf.md">ASRF</a> </td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
   <tr>
     <td colspan="5" style="font-weight:bold;">时空动作检测方法</td>
   </tr>

README_en.md

@@ -2,6 +2,12 @@
 
 # PaddleVideo
 
+## Update:
+
+- add skeleton-base action recognition model [CTR-GCN](./docs/en/model_zoo/recognition/ctrgcn.md).
+- add lite action recognition model [MoViNet](./docs/zh-CN/model_zoo/recognition/movinet.md).
+- add temporal segment model [MS-TCN](./docs/zh-CN/model_zoo/segmentation/mstcn.md), [ASRF](./docs/zh-CN/model_zoo/segmentation/asrf.md).
+
 ​ 💖 **Welcome to scan the code and join the group discussion** 💖
 
 <div align="center">
@@ -42,15 +48,15 @@ PaddleVideo is a toolset for video tasks prepared for the industry and academia.
     <td><a href="./docs/en/model_zoo/recognition/timesformer.md">TimeSformer</a> (Transformer')</td>
     <td><a href="./docs/en/model_zoo/recognition/videoswin.md">VideoSwin</a> (Transformer’)</td>
     <td><a href="./docs/en/model_zoo/recognition/attention_lstm.md">AttentionLSTM</a> (RNN')</td>
-    <td></td>
+    <td><a href="./docs/en/model_zoo/recognition/movinet.md">MoViNet</a> (Lite‘)</td>
   </tr>
   <tr>
     <td colspan="5" style="font-weight:bold;">Skeleton based action recognition</td>
   </tr>
   <tr>
     <td><a href="./docs/en/model_zoo/recognition/stgcn.md">ST-GCN</a> (Custom’)</td>
     <td><a href="./docs/en/model_zoo/recognition/agcn.md">AGCN</a> (Adaptive')</td>
-    <td></td>
+    <td><a href="./docs/en/model_zoo/recognition/ctrgcn.md">CTR-GCN</a> (GCN‘)</td>
     <td></td>
     <td></td>
   </tr>
@@ -64,6 +70,16 @@ PaddleVideo is a toolset for video tasks prepared for the industry and academia.
     <td></td>
     <td></td>
   </tr>
+  <tr>
+    <td colspan="5" style="font-weight:bold;">temporal segment</td>
+  </tr>
+  <tr>
+    <td><a href="./docs/en/model_zoo/segmentation/mstcn.md">MS-TCN</a> </td>
+    <td><a href="./docs/en/model_zoo/segmentation/asrf.md">ASRF</a> </td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
   <tr>
     <td colspan="5" style="font-weight:bold;">Spatio-temporal motion detection method</td>
   </tr>

applications/Ma-Net/README.md

@@ -28,7 +28,10 @@ This code currently supports model test and model training on DAVIS  dataset,  a
 Please refer to DAVIS data download and preparation doc [DAVIS-data](dataloaders/DAVIS2017.md)
 
 ## Train and Test
-
+- You can download [pertained model for stage1](https://videotag.bj.bcebos.com/PaddleVideo-release2.2/DeeplabV3_coco.pdparams) decompress it for stage1 training。
+  
+- You can download [trained model of stage1](https://videotag.bj.bcebos.com/PaddleVideo-release2.2/MaNet_davis2017_stage1.pdparams) decompress it for stage2 training directly skipping stage1 training。
+  
 ```
 sh run_local.sh
 ```

applications/Ma-Net/README_cn.md

@@ -28,11 +28,14 @@ DAVIS数据下载及准备请参考[DAVIS2017数据准备](dataloaders/DAVIS2017
 
 
 ## 模型训练与测试
-
+- 您可以下载[paddle版本的stage1预训练模型](https://videotag.bj.bcebos.com/PaddleVideo-release2.2/DeeplabV3_coco.pdparams) 解压缩它以用于训练的第一阶段。
+  
+- 您可以下载[stage1训练结果模型](https://videotag.bj.bcebos.com/PaddleVideo-release2.2/MaNet_davis2017_stage1.pdparams) 解压缩它以直接训练的第二阶段跳过第一阶段的训练。
+  
   ```bash
-  sh run_local.sh
+  sh run.sh
   ```
-
+  
 - 您可以下载[我们的模型](https://videotag.bj.bcebos.com/PaddleVideo-release2.2/MANet_davis2017.pdparams) 解压缩它以用于测试。