HRNet 训练自定义数据集

在这里插入图片描述

基于 HRNet 训练人脸特征点数据集.

INSTALL

conda create -n openmmlab python=3.8
# torch gpu```
pip install torch==1.12.1+cu102 torchvision==0.13.1+cu102 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu102
# 
pip install -U openmim
mim install mmcv-full

# install MMPose.
git clone https://github.com/open-mmlab/mmpose.git
cd mmpose
pip install -r requirements.txt
pip install -v -e .

Train custom dataset

添加 configs/__base__/dataset/custom.py , 配置特征点信息 (我的数据是 37个人脸特征点)

dataset_info = dict(
    dataset_name='landmark',
    paper_info=dict(
        author='Wu, Wayne and Qian, Chen and Yang, Shuo and Wang, '
               'Quan and Cai, Yici and Zhou, Qiang',
        title='Look at boundary: A boundary-aware face alignment algorithm',
        container='Proceedings of the IEEE conference on computer '
                  'vision and pattern recognition',
        year='2018',
        homepage='https://wywu.github.io/projects/LAB/WFLW.html',
    ),
    keypoint_info={
      
      
        0:
            dict(
                name='left_eye_left_corner', id=0, color=[255, 255, 255], type='', swap='right_eye_right_corner'),
        1:
            dict(
                name='left_eye_top_1', id=1, color=[255, 255, 255], type='', swap='right_eye_top_4'),
        2:
            dict(
                name='left_eye_top_2', id=2, color=[255, 255, 255], type='', swap='right_eye_top_3'),
        3:
            dict(
                name='left_eye_top_3', id=3, color=[255, 255, 255], type='', swap='right_eye_top_2'),
        4:
            dict(
                name='left_eye_top_4', id=4, color=[255, 255, 255], type='', swap='right_eye_top_1'),
        5:
            dict(
                name='left_eye_right_corner', id=5, color=[255, 255, 255], type='', swap='right_eye_left_corner'),
        6:
            dict(
                name='left_eye_bottom_1', id=6, color=[255, 255, 255], type='', swap='right_eye_bottom_4'),
        7:
            dict(
                name='left_eye_bottom_2', id=7, color=[255, 255, 255], type='', swap='right_eye_bottom_3'),
        8:
            dict(
                name='left_eye_bottom_3', id=8, color=[255, 255, 255], type='', swap='right_eye_bottom_2'),
        9:
            dict(
                name='left_eye_bottom_4', id=9, color=[255, 255, 255], type='', swap='right_eye_bottom_1'),
        10:
            dict(
                name='right_eye_left_corner',
                id=10,
                color=[255, 255, 255],
                type='',
                swap='left_eye_right_corner'),
        11:
            dict(
                name='right_eye_top_1',
                id=11,
                color=[255, 255, 255],
                type='',
                swap='left_eye_top_4'),
        12:
            dict(
                name='right_eye_top_2',
                id=12,
                color=[255, 255, 255],
                type='',
                swap='left_eye_top_3'),
        13:
            dict(
                name='right_eye_top_3',
                id=13,
                color=[255, 255, 255],
                type='',
                swap='left_eye_top_2'),
        14:
            dict(
                name='right_eye_top_4',
                id=14,
                color=[255, 255, 255],
                type='',
                swap='left_eye_top_1'),
        15:
            dict(
                name='right_eye_right_corner',
                id=15,
                color=[255, 255, 255],
                type='',
                swap='left_eye_left_corner'),
        16:
            dict(name='right_eye_bottom_1', id=16, color=[255, 255, 255], type='', swap='left_eye_bottom_4'),
        17:
            dict(
                name='right_eye_bottom_2',
                id=17,
                color=[255, 255, 255],
                type='',
                swap='left_eye_bottom_3'),
        18:
            dict(
                name='right_eye_bottom_3',
                id=18,
                color=[255, 255, 255],
                type='',
                swap='left_eye_bottom_2'),
        19:
            dict(
                name='right_eye_bottom_4',
                id=19,
                color=[255, 255, 255],
                type='',
                swap='left_eye_bottom_1'),
        20:
            dict(
                name='left_eyebrow_left_corner',
                id=20,
                color=[255, 255, 255],
                type='',
                swap='right_eyebrow_right_corner'),
        21:
            dict(
                name='left_eyebrow_right_corner',
                id=21,
                color=[255, 255, 255],
                type='',
                swap='right_eyebrow_left_corner'),
        22:
            dict(
                name='right_eyebrow_left_corner',
                id=22,
                color=[255, 255, 255],
                type='',
                swap='left_eyebrow_right_corner'),
        23:
            dict(
                name='right_eyebrow_right_corner',
                id=23,
                color=[255, 255, 255],
                type='',
                swap='left_eyebrow_left_corner'),
        24:
            dict(
                name='nose_left_corner',
                id=24,
                color=[255, 255, 255],
                type='',
                swap='nose_right_corner'),
        25:
            dict(
                name='nose_right_corner',
                id=25,
                color=[255, 255, 255],
                type='',
                swap='nose_left_corner'),
        26:
            dict(
                name='mouth_left_corner',
                id=26,
                color=[255, 255, 255],
                type='',
                swap='mouth_right_corner'),
        27:
            dict(
                name='mouth_top_1', id=27, color=[255, 255, 255], type='',
                swap='mouth_top_3'),
        28:
            dict(
                name='mouth_top_2', id=28, color=[255, 255, 255], type='',
                swap='mouth_top_2'),
        29:
            dict(
                name='mouth_top_3', id=29, color=[255, 255, 255], type='',
                swap='mouth_top_1'),
        30:
            dict(
                name='mouth_right_corner', id=30, color=[255, 255, 255], type='',
                swap='mouth_left_corner'),
        31:
            dict(
                name='mouth_bottom_1', id=31, color=[255, 255, 255], type='',
                swap='mouth_bottom_3'),
        32:
            dict(
                name='mouth_bottom_2', id=32, color=[255, 255, 255], type='',
                swap='mouth_bottom_2'),
        33:
            dict(
                name='mouth_bottom_3', id=33, color=[255, 255, 255], type='',
                swap='mouth_bottom_1'),
        34:
            dict(
                name='chin_left', id=34, color=[255, 255, 255], type='',
                swap='chin_right'),
        35:
            dict(
                name='chin_mid', id=35, color=[255, 255, 255], type='',
                swap='chin_mid'),
        36:
            dict(
                name='chin_right', id=36, color=[255, 255, 255], type='',
                swap='chin_left')
    },
    skeleton_info={
      
      },
    joint_weights=[1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5,
                   1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
    sigmas=[])

创建 mmpose/datasets/datasets/face/face_custom_dataset.py, 这里起名叫 FaceCustomDataset ,参考其他数据集，然后根据自己数据集修改就行.

init.py 增加 FaceCustomDataset

from .face_custom_dataset import FaceCustomDataset  
__all__ = [  
    'Face300WDataset', 'FaceAFLWDataset', 'FaceWFLWDataset', 'FaceCOFWDataset',  
    'FaceCocoWholeBodyDataset', 'FaceCustomDataset'  
]

创建 configs/custom/litehrnet_w18_face_256x192.py

# 修改 _base_ 改成 custom.py
_base_ = [  
    '../_base_/default_runtime.py',  
    '../_base_/datasets/custom.py'  
]
	...
	# 数据集配置
	data = dict(  
	    samples_per_gpu=64,  
	    workers_per_gpu=4,  
	    val_dataloader=dict(samples_per_gpu=32),  
	    test_dataloader=dict(samples_per_gpu=32),  
	    train=dict(  
	       # 设置我们自定义数据集
	        type='FaceCustomDataset',  
	        ann_file=f'{
        
        data_root}/annotations/face_landmark_train.json',  
	        img_prefix=f'{
        
        data_root}/images/',  
	        data_cfg=data_cfg,  
	        pipeline=train_pipeline,  
	        dataset_info={
      
      {
      
      _base_.dataset_info}}),  
	    val=dict(  
	        type='FaceCustomDataset',  
	        ann_file=f'{
        
        data_root}/annotations/face_landmark_val.json',  
	        img_prefix=f'{
        
        data_root}/images/',  
	        data_cfg=data_cfg,  
	        pipeline=val_pipeline,  
	        dataset_info={
      
      {
      
      _base_.dataset_info}}),  
	    test=dict(  
	        type='FaceCustomDataset',  
	        ann_file=f'{
        
        data_root}/annotations/face_landmarks_wflw_test.json',  
	        img_prefix=f'{
        
        data_root}/images/',  
	        data_cfg=data_cfg,  
	        pipeline=test_pipeline,  
	        dataset_info={
      
      {
      
      _base_.dataset_info}}),  
	)
	```

Train

# signle gpu
python .tools/train.py "./configs/custom/litehrnet_w18_face_256x192.py" -work-dir "./tools/runs/models"
# mul gpu
./tools/dist_train.sh "./configs/custom/litehrnet_w18_face_256x192.py" 4 --work-dir "./tools/runs/ld_37"

在这里插入图片描述
6. 验证
python demo/face_img_demo.py ./configs/custom/litehrnet_w18_face_256x192.py tools/runs/ld_37/best_NME_epoch_1660.pth --img-root tests/data/ld_37/ --img 1.jpg --out-img-root vis_results

HRNet 训练自定义数据集

基于 HRNet 训练人脸特征点数据集.

INSTALL

Train custom dataset

参考

猜你喜欢