easy_ViTPose/configs/ViTPose_common.py

# Common configuration
optimizer = dict(type='AdamW', lr=1e-3, betas=(0.9, 0.999), weight_decay=0.1,
                 constructor='LayerDecayOptimizerConstructor', 
                 paramwise_cfg=dict(
                                    num_layers=12, 
                                    layer_decay_rate=1 - 2e-4,
                                    custom_keys={
                                            'bias': dict(decay_multi=0.),
                                            'pos_embed': dict(decay_mult=0.),
                                            'relative_position_bias_table': dict(decay_mult=0.),
                                            'norm': dict(decay_mult=0.)
                                            }
                                    )
                )

optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))

# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=300,
    warmup_ratio=0.001,
    step=[3])

total_epochs = 4
target_type = 'GaussianHeatmap'

data_cfg = dict(
    image_size=[192, 256],
    heatmap_size=[48, 64],
    soft_nms=False,
    nms_thr=1.0,
    oks_thr=0.9,
    vis_thr=0.2,
    use_gt_bbox=False,
    det_bbox_thr=0.0,
    bbox_file='data/coco/person_detection_results/'
    'COCO_val2017_detections_AP_H_56_person.json',
)

data_root = '/home/adryw/dataset/COCO17'
data = dict(
    samples_per_gpu=64,
    workers_per_gpu=6,
    val_dataloader=dict(samples_per_gpu=128),
    test_dataloader=dict(samples_per_gpu=128),
    train=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
        img_prefix=f'{data_root}/train2017/',
        data_cfg=data_cfg),
    val=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
        img_prefix=f'{data_root}/val2017/',
        data_cfg=data_cfg),
    test=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
        img_prefix=f'{data_root}/val2017/',
        data_cfg=data_cfg)
)

model_small = dict(
    type='TopDown',
    pretrained=None,
    backbone=dict(
        type='ViT',
        img_size=(256, 192),
        patch_size=16,
        embed_dim=384,
        depth=12,
        num_heads=12,
        ratio=1,
        use_checkpoint=False,
        mlp_ratio=4,
        qkv_bias=True,
        drop_path_rate=0.1,
    ),
    keypoint_head=dict(
        type='TopdownHeatmapSimpleHead',
        in_channels=384,
        num_deconv_layers=2,
        num_deconv_filters=(256, 256),
        num_deconv_kernels=(4, 4),
        extra=dict(final_conv_kernel=1, ),
        loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=False,
        target_type=target_type,
        modulate_kernel=11,
        use_udp=True))

model_base = dict(
    type='TopDown',
    pretrained=None,
    backbone=dict(
        type='ViT',
        img_size=(256, 192),
        patch_size=16,
        embed_dim=768,
        depth=12,
        num_heads=12,
        ratio=1,
        use_checkpoint=False,
        mlp_ratio=4,
        qkv_bias=True,
        drop_path_rate=0.3,
    ),
    keypoint_head=dict(
        type='TopdownHeatmapSimpleHead',
        in_channels=768,
        num_deconv_layers=2,
        num_deconv_filters=(256, 256),
        num_deconv_kernels=(4, 4),
        extra=dict(final_conv_kernel=1, ),
        loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=False,
        target_type=target_type,
        modulate_kernel=11,
        use_udp=True))

model_large = dict(
    type='TopDown',
    pretrained=None,
    backbone=dict(
        type='ViT',
        img_size=(256, 192),
        patch_size=16,
        embed_dim=1024,
        depth=24,
        num_heads=16,
        ratio=1,
        use_checkpoint=False,
        mlp_ratio=4,
        qkv_bias=True,
        drop_path_rate=0.5,
    ),
    keypoint_head=dict(
        type='TopdownHeatmapSimpleHead',
        in_channels=1024,
        num_deconv_layers=2,
        num_deconv_filters=(256, 256),
        num_deconv_kernels=(4, 4),
        extra=dict(final_conv_kernel=1, ),
        loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=False,
        target_type=target_type,
        modulate_kernel=11,
        use_udp=True))

model_huge = dict(
    type='TopDown',
    pretrained=None,
    backbone=dict(
        type='ViT',
        img_size=(256, 192),
        patch_size=16,
        embed_dim=1280,
        depth=32,
        num_heads=16,
        ratio=1,
        use_checkpoint=False,
        mlp_ratio=4,
        qkv_bias=True,
        drop_path_rate=0.55,
    ),
    keypoint_head=dict(
        type='TopdownHeatmapSimpleHead',
        in_channels=1280,
        num_deconv_layers=2,
        num_deconv_filters=(256, 256),
        num_deconv_kernels=(4, 4),
        extra=dict(final_conv_kernel=1, ),
        loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=False,
        target_type=target_type,
        modulate_kernel=11,
        use_udp=True))