From 6e043ebc3ff9e43715ecb5d4cce9b0a020b952db Mon Sep 17 00:00:00 2001 From: Tau-J <674106399@qq.com> Date: Mon, 10 Oct 2022 15:15:54 +0800 Subject: [PATCH 1/2] add algorithm readme --- .../integral_regression/README.md | 15 ++++++++++++++ configs/body_2d_keypoint/simcc/README.md | 20 +++++++++++++++++++ docs/src/papers/algorithms/simcc.md | 2 +- 3 files changed, 36 insertions(+), 1 deletion(-) create mode 100644 configs/body_2d_keypoint/integral_regression/README.md create mode 100644 configs/body_2d_keypoint/simcc/README.md diff --git a/configs/body_2d_keypoint/integral_regression/README.md b/configs/body_2d_keypoint/integral_regression/README.md new file mode 100644 index 0000000000..d60eaa1a57 --- /dev/null +++ b/configs/body_2d_keypoint/integral_regression/README.md @@ -0,0 +1,15 @@ +# Top-down integral-regression-based pose estimation + +Top-down methods divide the task into two stages: object detection, followed by single-object pose estimation given object bounding boxes. At the 2nd stage, integral regression based methods use a simple integral operation relates and unifies the heatmap and joint regression differentiably, thus obtain the keypoint coordinates given the features extracted from the bounding box area, following the paradigm introduced in [Integral Human Pose Regression](https://arxiv.org/abs/1711.08229). + +## Results and Models + +### COCO Dataset + +Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset + +| Model | Input Size | AP | AR | Details and Download | +| :------------------: | :--------: | :---: | :---: | :---------------------------------------------------: | +| ResNet-50+Debias-IPR | 256x256 | 0.675 | 0.765 | [resnet_debias_coco.md](./coco/resnet_debias_coco.md) | +| ResNet-50+DSNT | 256x256 | 0.674 | 0.764 | [resnet_dsnt_coco.md](./coco/resnet_dsnt_coco.md) | +| ResNet-50+IPR | 256x256 | 0.633 | 0.730 | [resnet_ipr_coco.md](./coco/resnet_ipr_coco.md) | diff --git a/configs/body_2d_keypoint/simcc/README.md b/configs/body_2d_keypoint/simcc/README.md new file mode 100644 index 0000000000..6148c18bf5 --- /dev/null +++ b/configs/body_2d_keypoint/simcc/README.md @@ -0,0 +1,20 @@ +# Top-down SimCC-based pose estimation + +Top-down methods divide the task into two stages: object detection, followed by single-object pose estimation given object bounding boxes. At the 2nd stage, SimCC based methods reformulate human pose estimation as two classification tasks for horizontal and vertical coordinates, and uniformly divide each pixel into several bins, thus obtain the keypoint coordinates given the features extracted from the bounding box area, following the paradigm introduced in [SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation](https://arxiv.org/abs/2107.03332). + +
+ +
+ +## Results and Models + +### COCO Dataset + +Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset + +| Model | Input Size | AP | AR | Details and Download | +| :---------------------------: | :--------: | :---: | :---: | :-----------------------------------------------: | +| ResNet-50+SimCC | 384x288 | 0.735 | 0.790 | [resnet_coco.md](./coco/resnet_coco.md) | +| ResNet-50+SimCC | 256x192 | 0.721 | 0.781 | [resnet_coco.md](./coco/resnet_coco.md) | +| S-ViPNAS-MobileNet-V3+SimCC | 256x192 | 0.695 | 0.755 | [vipnas_coco.md](./coco/vipnas_coco.md) | +| MobileNet-V2+SimCC(wo/deconv) | 256x192 | 0.620 | 0.678 | [mobilenetv2_coco.md](./coco/mobilenetv2_coco.md) | diff --git a/docs/src/papers/algorithms/simcc.md b/docs/src/papers/algorithms/simcc.md index 3a6f0816b6..9a00d229f3 100644 --- a/docs/src/papers/algorithms/simcc.md +++ b/docs/src/papers/algorithms/simcc.md @@ -1,4 +1,4 @@ -# Simple Coordinate Classification Perspective for Human Pose Estimation +# SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation From 75a215f84462b1f3396e3422b78c8d24ebd7fb46 Mon Sep 17 00:00:00 2001 From: Tau-J <674106399@qq.com> Date: Mon, 10 Oct 2022 17:32:01 +0800 Subject: [PATCH 2/2] update ckpt --- configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md b/configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md index fa4d5e0d39..42438774ba 100644 --- a/configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md +++ b/configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md @@ -52,4 +52,4 @@ Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 da | Arch | Input Size | AP | AP50 | AP75 | AR | AR50 | ckpt | log | | :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: | -| [simcc_mobilenetv2_wo_deconv](/configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py) | 256x192 | 0.620 | 0.855 | 0.697 | 0.678 | 0.902 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-e0cc028d_20220922.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-e0cc028d_20220922.log.json) | +| [simcc_mobilenetv2_wo_deconv](/configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py) | 256x192 | 0.620 | 0.855 | 0.697 | 0.678 | 0.902 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-4b0703bb_20221010.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-4b0703bb_20221010.log.json) |