update hrnetv2 ckpt and configs (#980)

open-mmlab · Oct 20, 2021 · 0069aff · 0069aff
1 parent a0d513b
commit 0069aff
Show file tree

Hide file tree

Showing 8 changed files with 46 additions and 9 deletions.
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_300w.md b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_300w.md
@@ -35,4 +35,10 @@
 
 </details>
 
-Checkpoints will be revealed after mmpose reaches 1k star :D
+Results on 300W dataset
+
+The model is trained on 300W train.
+
+| Arch  | Input Size | NME<sub>*common*</sub> | NME<sub>*challenge*</sub> | NME<sub>*full*</sub> | NME<sub>*test*</sub> | ckpt | log |
+| :-----| :--------: | :------------------: | :------------------: | :--------------: |:-------------------------: |:---: | :---: |
+| [pose_hrnetv2_w18](/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_w18_300w_256x256.py)  | 256x256 | 2.86 | 5.45 | 3.37  | 3.97  | [ckpt](https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_300w_256x256-eea53406_20211019.pth) | [log](https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_300w_256x256_20211019.log.json) |
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_300w.yml b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_300w.yml
@@ -7,4 +7,18 @@ Collections:
     Title: Deep High-Resolution Representation Learning for Visual Recognition
     URL: https://ieeexplore.ieee.org/abstract/document/9052469/
   README: configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_300w.md
-Models: []
+Models:
+- Config: configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_w18_300w_256x256.py
+  In Collection: Face 2D Keypoint topdown_heatmap 300w
+  Metadata:
+    Training Data: 300W
+  Name: topdown_heatmap_hrnetv2_w18_300w_256x256
+  Results:
+  - Dataset: 300W
+    Metrics:
+      NME challenge: 5.45
+      NME common: 2.86
+      NME full: 3.37
+      NME test: 3.97
+    Task: Face 2D Keypoint
+  Weights: https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_300w_256x256-eea53406_20211019.pth
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_w18_300w_256x256.py b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/300w/hrnetv2_w18_300w_256x256.py
@@ -106,7 +106,7 @@
         type='NormalizeTensor',
         mean=[0.485, 0.456, 0.406],
         std=[0.229, 0.224, 0.225]),
-    dict(type='TopDownGenerateTarget', sigma=2),
+    dict(type='TopDownGenerateTarget', sigma=1.5),
     dict(
         type='Collect',
         keys=['img', 'target', 'target_weight'],

diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_dark_aflw.md b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_dark_aflw.md
@@ -57,4 +57,4 @@ The model is trained on AFLW train and evaluated on AFLW full and frontal.
 
 | Arch  | Input Size | NME<sub>*full*</sub> | NME<sub>*frontal*</sub>  | ckpt | log |
 | :-------------- | :-----------: | :------: | :------: |:------: |:------: |
-| [pose_hrnetv2_w18_dark](/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_w18_aflw_256x256_dark.py)  | 256x256 | 1.41 | 1.27 | [ckpt](https://download.openmmlab.com/mmpose/face/darkpose/hrnetv2_w18_aflw_256x256_dark-219606c0_20210125.pth) | [log](https://download.openmmlab.com/mmpose/face/darkpose/hrnetv2_w18_aflw_256x256_dark_20210125.log.json) |
+| [pose_hrnetv2_w18_dark](/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_w18_aflw_256x256_dark.py)  | 256x256 | 1.34 | 1.20 | [ckpt](https://download.openmmlab.com/mmpose/face/darkpose/hrnetv2_w18_aflw_256x256_dark-219606c0_20210125.pth) | [log](https://download.openmmlab.com/mmpose/face/darkpose/hrnetv2_w18_aflw_256x256_dark_20210125.log.json) |
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_dark_aflw.yml b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/aflw/hrnetv2_dark_aflw.yml
@@ -17,7 +17,7 @@ Models:
   Results:
   - Dataset: AFLW
     Metrics:
-      NME frontal: 1.27
-      NME full: 1.41
+      NME frontal: 1.2
+      NME full: 1.34
     Task: Face 2D Keypoint
   Weights: https://download.openmmlab.com/mmpose/face/darkpose/hrnetv2_w18_aflw_256x256_dark-219606c0_20210125.pth
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_cofw.md b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_cofw.md
@@ -33,4 +33,10 @@
 
 </details>
 
-Checkpoints will be revealed after mmpose reaches 1k star :D
+Results on COFW dataset
+
+The model is trained on COFW train.
+
+| Arch  | Input Size | NME | ckpt | log |
+| :-----| :--------: | :----: |:---: | :---: |
+| [pose_hrnetv2_w18](/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_w18_cofw_256x256.py)  | 256x256 |  3.40 | [ckpt](https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_cofw_256x256-49243ab8_20211019.pth) | [log](https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_cofw_256x256_20211019.log.json) |
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_cofw.yml b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_cofw.yml
@@ -7,4 +7,15 @@ Collections:
     Title: Deep High-Resolution Representation Learning for Visual Recognition
     URL: https://ieeexplore.ieee.org/abstract/document/9052469/
   README: configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_cofw.md
-Models: []
+Models:
+- Config: configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_w18_cofw_256x256.py
+  In Collection: Face 2D Keypoint topdown_heatmap cofw
+  Metadata:
+    Training Data: COFW
+  Name: topdown_heatmap_hrnetv2_w18_cofw_256x256
+  Results:
+  - Dataset: COFW
+    Metrics:
+      NME: 3.4
+    Task: Face 2D Keypoint
+  Weights: https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_cofw_256x256-49243ab8_20211019.pth
diff --git a/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_w18_cofw_256x256.py b/configs/face/2d_kpt_sview_rgb_img/topdown_heatmap/cofw/hrnetv2_w18_cofw_256x256.py
@@ -106,7 +106,7 @@
         type='NormalizeTensor',
         mean=[0.485, 0.456, 0.406],
         std=[0.229, 0.224, 0.225]),
-    dict(type='TopDownGenerateTarget', sigma=2),
+    dict(type='TopDownGenerateTarget', sigma=1.5),
     dict(
         type='Collect',
         keys=['img', 'target', 'target_weight'],