update rtmw

open-mmlab · Sep 26, 2023 · 4ae6a9a · 4ae6a9a
1 parent f3fdb86
commit 4ae6a9a
Show file tree

Hide file tree

Showing 3 changed files with 21 additions and 19 deletions.
diff --git a/projects/rtmpose/README.md b/projects/rtmpose/README.md
@@ -46,6 +46,7 @@ ______________________________________________________________________
 
 - Sep. 2023:
   - Add RTMW models trained on combined datasets. The alpha version of RTMW-x model achieves 70.2 mAP on COCO-Wholebody val set. The technical report will be released soon.
+  - Add YOLOX and RTMDet models trained on HumanArt dataset.
 - Aug. 2023:
   - Support distilled 133-keypoint WholeBody models powered by [DWPose](https://github.com/IDEA-Research/DWPose/tree/main).
     - You can try DWPose/RTMPose with [sd-webui-controlnet](https://github.com/Mikubill/sd-webui-controlnet) now! Just update your sd-webui-controlnet >= v1.1237, then choose `dw_openpose_full` as preprocessor.
@@ -221,15 +222,6 @@ Feel free to join our community group for more help:
 - RTMPose for Human-Centric Artificial Scenes is supported by  [Human-Art](https://github.com/IDEA-Research/HumanArt)
 - <img src="https://github.com/open-mmlab/mmpose/assets/13503330/685bc610-dd9e-4e6f-9c41-dbc8220584f4" height="300px">
 
-Pose Estimators:
-
-|                                     Config                                      | Input Size | AP<sup><br>(Human-Art GT) | Params<sup><br>(M) | FLOPS<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) | ncnn-FP16-Latency<sup><br>(ms)<sup><br>(Snapdragon 865) |                                                                                                                                   Download                                                                                                                                   |
-| :-----------------------------------------------------------------------------: | :--------: | :-----------------------: | :----------------: | :---------------: | :-----------------------------------------: | :------------------------------------------------: | :-----------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
-| [RTMPose-t\*](./rtmpose/body_2d_keypoint/rtmpose-t_8xb256-420e_coco-256x192.py) |  256x192   |           65.5            |        3.34        |       0.36        |                    3.20                     |                        1.06                        |                          9.02                           | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.zip) |
-| [RTMPose-s\*](./rtmpose/body_2d_keypoint/rtmpose-s_8xb256-420e_coco-256x192.py) |  256x192   |           69.8            |        5.47        |       0.68        |                    4.48                     |                        1.39                        |                          13.89                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.zip) |
-| [RTMPose-m\*](./rtmpose/body_2d_keypoint/rtmpose-m_8xb256-420e_coco-256x192.py) |  256x192   |           72.8            |       13.59        |       1.93        |                    11.06                    |                        2.29                        |                          26.44                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.zip) |
-| [RTMPose-l\*](./rtmpose/body_2d_keypoint/rtmpose-l_8xb256-420e_coco-256x192.py) |  256x192   |           75.3            |       27.66        |       4.16        |                    18.85                    |                        3.46                        |                          45.37                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.zip) |
-
 Detetors:
 
 |       Detection Config        | Input Size | Model AP<sup><br>(OneHand10K) | Flops<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) |        Download        |
@@ -243,6 +235,15 @@ Detetors:
 | [YOLOX-l](./yolox/humanart/yolox_l_8xb8-300e_humanart.py) |  640x640   |             60.2              |         -         |                      -                      |                         -                          | [Det Model](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/yolox_l_8xb8-300e_humanart-ce1d7a62.pth) |
 | [YOLOX-x](./yolox/humanart/yolox_x_8xb8-300e_humanart.py) |  640x640   |             61.3              |         -         |                      -                      |                         -                          | [Det Model](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/yolox_x_8xb8-300e_humanart-a39d44ed.pth) |
 
+Pose Estimators:
+
+|                                     Config                                      | Input Size | AP<sup><br>(Human-Art GT) | Params<sup><br>(M) | FLOPS<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) | ncnn-FP16-Latency<sup><br>(ms)<sup><br>(Snapdragon 865) |                                                                                                                                   Download                                                                                                                                   |
+| :-----------------------------------------------------------------------------: | :--------: | :-----------------------: | :----------------: | :---------------: | :-----------------------------------------: | :------------------------------------------------: | :-----------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
+| [RTMPose-t\*](./rtmpose/body_2d_keypoint/rtmpose-t_8xb256-420e_coco-256x192.py) |  256x192   |           65.5            |        3.34        |       0.36        |                    3.20                     |                        1.06                        |                          9.02                           | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.zip) |
+| [RTMPose-s\*](./rtmpose/body_2d_keypoint/rtmpose-s_8xb256-420e_coco-256x192.py) |  256x192   |           69.8            |        5.47        |       0.68        |                    4.48                     |                        1.39                        |                          13.89                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.zip) |
+| [RTMPose-m\*](./rtmpose/body_2d_keypoint/rtmpose-m_8xb256-420e_coco-256x192.py) |  256x192   |           72.8            |       13.59        |       1.93        |                    11.06                    |                        2.29                        |                          26.44                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.zip) |
+| [RTMPose-l\*](./rtmpose/body_2d_keypoint/rtmpose-l_8xb256-420e_coco-256x192.py) |  256x192   |           75.3            |       27.66        |       4.16        |                    18.85                    |                        3.46                        |                          45.37                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.zip) |
+
 </details>
 
 #### 26 Keypoints

diff --git a/projects/rtmpose/README_CN.md b/projects/rtmpose/README_CN.md
@@ -42,6 +42,7 @@ ______________________________________________________________________
 
 - 2023 年 9 月：
   - 发布混合数据集上训练的 RTMW 模型。Alpha 版本的 RTMW-x 在 COCO-Wholebody 验证集上取得了 70.2 mAP。技术报告正在撰写中。
+  - 增加 HumanArt 上训练的 YOLOX 和 RTMDet 模型。
 - 2023 年 8 月：
   - 支持基于 RTMPose 模型蒸馏的 133 点 WholeBody 模型（由 [DWPose](https://github.com/IDEA-Research/DWPose/tree/main) 提供）。
     - 你可以在 [sd-webui-controlnet](https://github.com/Mikubill/sd-webui-controlnet) 中使用 DWPose/RTMPose 作为姿态估计后端进行人物图像生成。升级 sd-webui-controlnet >= v1.1237 并选择 `dw_openpose_full` 即可使用。
@@ -212,15 +213,6 @@ RTMPose 是一个长期优化迭代的项目，致力于业务场景下的高性
 - 面向艺术图片的人体姿态估计 RTMPose 模型由 [Human-Art](https://github.com/IDEA-Research/HumanArt) 提供。
 - <img src="https://github.com/open-mmlab/mmpose/assets/13503330/685bc610-dd9e-4e6f-9c41-dbc8220584f4" height="300px">
 
-人体姿态估计模型：
-
-|                                     Config                                      | Input Size | AP<sup><br>(Human-Art GT) | Params<sup><br>(M) | FLOPS<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) | ncnn-FP16-Latency<sup><br>(ms)<sup><br>(Snapdragon 865) |                                                                                                                                   Download                                                                                                                                   |
-| :-----------------------------------------------------------------------------: | :--------: | :-----------------------: | :----------------: | :---------------: | :-----------------------------------------: | :------------------------------------------------: | :-----------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
-| [RTMPose-t\*](./rtmpose/body_2d_keypoint/rtmpose-t_8xb256-420e_coco-256x192.py) |  256x192   |           65.5            |        3.34        |       0.36        |                    3.20                     |                        1.06                        |                          9.02                           | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.zip) |
-| [RTMPose-s\*](./rtmpose/body_2d_keypoint/rtmpose-s_8xb256-420e_coco-256x192.py) |  256x192   |           69.8            |        5.47        |       0.68        |                    4.48                     |                        1.39                        |                          13.89                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.zip) |
-| [RTMPose-m\*](./rtmpose/body_2d_keypoint/rtmpose-m_8xb256-420e_coco-256x192.py) |  256x192   |           72.8            |       13.59        |       1.93        |                    11.06                    |                        2.29                        |                          26.44                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.zip) |
-| [RTMPose-l\*](./rtmpose/body_2d_keypoint/rtmpose-l_8xb256-420e_coco-256x192.py) |  256x192   |           75.3            |       27.66        |       4.16        |                    18.85                    |                        3.46                        |                          45.37                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.zip) |
-
 人体检测模型：
 
 |       Detection Config        | Input Size | Model AP<sup><br>(OneHand10K) | Flops<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) |        Download        |
@@ -234,6 +226,15 @@ RTMPose 是一个长期优化迭代的项目，致力于业务场景下的高性
 | [YOLOX-l](./yolox/humanart/yolox_l_8xb8-300e_humanart.py) |  640x640   |             60.2              |         -         |                      -                      |                         -                          | [Det Model](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/yolox_l_8xb8-300e_humanart-ce1d7a62.pth) |
 | [YOLOX-x](./yolox/humanart/yolox_x_8xb8-300e_humanart.py) |  640x640   |             61.3              |         -         |                      -                      |                         -                          | [Det Model](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/yolox_x_8xb8-300e_humanart-a39d44ed.pth) |
 
+人体姿态估计模型：
+
+|                                     Config                                      | Input Size | AP<sup><br>(Human-Art GT) | Params<sup><br>(M) | FLOPS<sup><br>(G) | ORT-Latency<sup><br>(ms)<sup><br>(i7-11700) | TRT-FP16-Latency<sup><br>(ms)<sup><br>(GTX 1660Ti) | ncnn-FP16-Latency<sup><br>(ms)<sup><br>(Snapdragon 865) |                                                                                                                                   Download                                                                                                                                   |
+| :-----------------------------------------------------------------------------: | :--------: | :-----------------------: | :----------------: | :---------------: | :-----------------------------------------: | :------------------------------------------------: | :-----------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
+| [RTMPose-t\*](./rtmpose/body_2d_keypoint/rtmpose-t_8xb256-420e_coco-256x192.py) |  256x192   |           65.5            |        3.34        |       0.36        |                    3.20                     |                        1.06                        |                          9.02                           | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.zip) |
+| [RTMPose-s\*](./rtmpose/body_2d_keypoint/rtmpose-s_8xb256-420e_coco-256x192.py) |  256x192   |           69.8            |        5.47        |       0.68        |                    4.48                     |                        1.39                        |                          13.89                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.zip) |
+| [RTMPose-m\*](./rtmpose/body_2d_keypoint/rtmpose-m_8xb256-420e_coco-256x192.py) |  256x192   |           72.8            |       13.59        |       1.93        |                    11.06                    |                        2.29                        |                          26.44                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.zip) |
+| [RTMPose-l\*](./rtmpose/body_2d_keypoint/rtmpose-l_8xb256-420e_coco-256x192.py) |  256x192   |           75.3            |       27.66        |       4.16        |                    18.85                    |                        3.46                        |                          45.37                          | [pth](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth)<br>[onnx](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.zip) |
+
 </details>
 
 #### 26 Keypoints

diff --git a/tools/misc/pth_transfer.py b/tools/misc/pth_transfer.py
@@ -6,7 +6,7 @@
 
 
 def change_model(args):
-    dis_model = torch.load(args.dis_path)
+    dis_model = torch.load(args.dis_path, map_location='cpu')
     all_name = []
     if args.two_dis:
         for name, v in dis_model['state_dict'].items():