From 361fba19fd2165d2de4a065804b8d2224adb19ef Mon Sep 17 00:00:00 2001
From: Xin Li <7219519+xin-li-67@users.noreply.github.com>
Date: Fri, 21 Apr 2023 20:33:24 +0800
Subject: [PATCH] [MMSIG-92] Integrate WFLW deeppose model to dev-1.x branch
(#2265)
---
.../topdown_regression/README.md | 17 +++
.../topdown_regression/wflw/resnet_wflw.md | 58 +++++++++
.../topdown_regression/wflw/resnet_wflw.yml | 15 +++
.../td-reg_res50_8x64e-210e_wflw-256x256.py | 122 ++++++++++++++++++
4 files changed, 212 insertions(+)
create mode 100644 configs/face_2d_keypoint/topdown_regression/README.md
create mode 100644 configs/face_2d_keypoint/topdown_regression/wflw/resnet_wflw.md
create mode 100644 configs/face_2d_keypoint/topdown_regression/wflw/resnet_wflw.yml
create mode 100644 configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py
diff --git a/configs/face_2d_keypoint/topdown_regression/README.md b/configs/face_2d_keypoint/topdown_regression/README.md
new file mode 100644
index 0000000000..030ee6b8fa
--- /dev/null
+++ b/configs/face_2d_keypoint/topdown_regression/README.md
@@ -0,0 +1,17 @@
+# Top-down 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, regression based methods directly regress the keypoint coordinates given the features extracted from the bounding box area, following the paradigm introduced in [Deeppose: Human pose estimation via deep neural networks](http://openaccess.thecvf.com/content_cvpr_2014/html/Toshev_DeepPose_Human_Pose_2014_CVPR_paper.html).
+
+
+
+
+DeepPose (CVPR'2014)
+
+```bibtex
+@inproceedings{toshev2014deeppose,
+ title={Deeppose: Human pose estimation via deep neural networks},
+ author={Toshev, Alexander and Szegedy, Christian},
+ booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
+ pages={1653--1660},
+ year={2014}
+}
+```
+
+
+
+
+
+
+ResNet (CVPR'2016)
+
+```bibtex
+@inproceedings{he2016deep,
+ title={Deep residual learning for image recognition},
+ author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
+ booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
+ pages={770--778},
+ year={2016}
+}
+```
+
+
+
+
+
+
+WFLW (CVPR'2018)
+
+```bibtex
+@inproceedings{wu2018look,
+ title={Look at boundary: A boundary-aware face alignment algorithm},
+ author={Wu, Wayne and Qian, Chen and Yang, Shuo and Wang, Quan and Cai, Yici and Zhou, Qiang},
+ booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
+ pages={2129--2138},
+ year={2018}
+}
+```
+
+
+
+Results on WFLW dataset
+
+The model is trained on WFLW train set.
+
+| Model | Input Size | NME | ckpt | log |
+| :-------------------------------------------------------------- | :--------: | :--: | :------------------------------------------------------------: | :-----------------------------------------------------------: |
+| [ResNet-50](/configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py) | 256x256 | 4.88 | [ckpt](https://download.openmmlab.com/mmpose/face/deeppose/deeppose_res50_wflw_256x256-92d0ba7f_20210303.pth) | [log](https://download.openmmlab.com/mmpose/face/deeppose/deeppose_res50_wflw_256x256_20210303.log.json) |
diff --git a/configs/face_2d_keypoint/topdown_regression/wflw/resnet_wflw.yml b/configs/face_2d_keypoint/topdown_regression/wflw/resnet_wflw.yml
new file mode 100644
index 0000000000..113cb33fc3
--- /dev/null
+++ b/configs/face_2d_keypoint/topdown_regression/wflw/resnet_wflw.yml
@@ -0,0 +1,15 @@
+Models:
+- Config: configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py
+ In Collection: ResNet
+ Metadata:
+ Architecture:
+ - DeepPose
+ - ResNet
+ Training Data: WFLW
+ Name: td-reg_res50_8x64e-210e_wflw-256x256
+ Results:
+ - Dataset: WFLW
+ Metrics:
+ NME: 4.88
+ Task: Face 2D Keypoint
+ Weights: https://download.openmmlab.com/mmpose/face/deeppose/deeppose_res50_wflw_256x256-92d0ba7f_20210303.pth
diff --git a/configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py b/configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py
new file mode 100644
index 0000000000..2742f497b8
--- /dev/null
+++ b/configs/face_2d_keypoint/topdown_regression/wflw/td-reg_res50_8x64e-210e_wflw-256x256.py
@@ -0,0 +1,122 @@
+_base_ = ['../../../_base_/default_runtime.py']
+
+# runtime
+train_cfg = dict(max_epochs=210, val_interval=10)
+
+# optimizer
+optim_wrapper = dict(optimizer=dict(
+ type='Adam',
+ lr=5e-4,
+))
+
+# learning policy
+param_scheduler = [
+ dict(
+ type='LinearLR', begin=0, end=500, start_factor=0.001,
+ by_epoch=False), # warm-up
+ dict(
+ type='MultiStepLR',
+ begin=0,
+ end=210,
+ milestones=[170, 200],
+ gamma=0.1,
+ by_epoch=True)
+]
+
+# automatically scaling LR based on the actual training batch size
+auto_scale_lr = dict(base_batch_size=512)
+
+# codec settings
+codec = dict(type='RegressionLabel', input_size=(256, 256))
+
+# model settings
+model = dict(
+ type='TopdownPoseEstimator',
+ data_preprocessor=dict(
+ type='PoseDataPreprocessor',
+ mean=[123.675, 116.28, 103.53],
+ std=[58.395, 57.12, 57.375],
+ bgr_to_rgb=True),
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'),
+ ),
+ neck=dict(type='GlobalAveragePooling'),
+ head=dict(
+ type='RegressionHead',
+ in_channels=2048,
+ num_joints=98,
+ loss=dict(type='SmoothL1Loss', use_target_weight=True),
+ decoder=codec),
+ train_cfg=dict(),
+ test_cfg=dict(
+ flip_test=True,
+ shift_coords=True,
+ ))
+
+# base dataset settings
+dataset_type = 'WFLWDataset'
+data_mode = 'topdown'
+data_root = 'data/wflw/'
+
+# pipelines
+train_pipeline = [
+ dict(type='LoadImage'),
+ dict(type='GetBBoxCenterScale'),
+ dict(type='RandomFlip', direction='horizontal'),
+ dict(
+ type='RandomBBoxTransform',
+ scale_factor=[0.75, 1.25],
+ rotate_factor=60),
+ dict(type='TopdownAffine', input_size=codec['input_size']),
+ dict(type='GenerateTarget', encoder=codec),
+ dict(type='PackPoseInputs')
+]
+val_pipeline = [
+ dict(type='LoadImage'),
+ dict(type='GetBBoxCenterScale'),
+ dict(type='TopdownAffine', input_size=codec['input_size']),
+ dict(type='PackPoseInputs')
+]
+
+# dataloaders
+train_dataloader = dict(
+ batch_size=64,
+ num_workers=2,
+ persistent_workers=True,
+ sampler=dict(type='DefaultSampler', shuffle=True),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ data_mode=data_mode,
+ ann_file='annotations/face_landmarks_wflw_train.json',
+ data_prefix=dict(img='images/'),
+ pipeline=train_pipeline,
+ ))
+val_dataloader = dict(
+ batch_size=32,
+ num_workers=2,
+ persistent_workers=True,
+ drop_last=False,
+ sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ data_mode=data_mode,
+ ann_file='annotations/face_landmarks_wflw_test.json',
+ data_prefix=dict(img='images/'),
+ test_mode=True,
+ pipeline=val_pipeline,
+ ))
+test_dataloader = val_dataloader
+
+# hooks
+default_hooks = dict(checkpoint=dict(save_best='NME', rule='less'))
+
+# evaluators
+val_evaluator = dict(
+ type='NME',
+ norm_mode='keypoint_distance',
+)
+test_evaluator = val_evaluator