mask_rcnn_dbswin-t_fpn_3x_nuim_cocopre.py

# _base_未使用
_base_ = [
    '../../_base_/models/mask_rcnn_r50_fpn.py',
    '../../_base_/datasets/nuim_instance.py',
    '../../_base_/schedules/mmdet_schedule_1x.py', '../_base_/default_runtime.py'
]

optimizer = dict(_delete_=True, type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.05,
                 paramwise_cfg=dict(custom_keys={'absolute_pos_embed': dict(decay_mult=0.),
                                                 'relative_position_bias_table': dict(decay_mult=0.),
                                                 'norm': dict(decay_mult=0.)}))
lr_config = dict(step=[27, 33])
runner = dict(type='EpochBasedRunner', max_epochs=36)
total_epochs = 36

# NOTE 使用了SwinTransformer的Parameter
pretrained = 'https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_tiny_patch4_window7_224.pth'

# NOTE backbone使用的是在SwinTransformer基础上的CSwinTransformer（CVPR 2022）
model = dict(
    pretrained = pretrained,  
    backbone=dict(
         _delete_=True,
        type='CBSwinTransformer',
        embed_dim=96,
        depths=[2, 2, 6, 2],
        num_heads=[3, 6, 12, 24],
        window_size=7,
        mlp_ratio=4.,
        qkv_bias=True,
        qk_scale=None,
        drop_rate=0.,
        attn_drop_rate=0.,
        drop_path_rate=0.2,
        ape=False,
        patch_norm=True,
        out_indices=(0, 1, 2, 3),
        use_checkpoint=True,
        ),
    neck=dict(
        in_channels=[96, 192, 384, 768]),
    roi_head=dict(
        bbox_head=dict(num_classes=10), mask_head=dict(num_classes=10)))

data = dict(
    samples_per_gpu=2,
    workers_per_gpu=2,)
# fp16 = dict(loss_scale=32.0)

# load_from未使用
load_from = 'work_dirs/mask_rcnn_cbv2_swin_tiny_patch4_window7_mstrain_480-800_adamw_3x_coco/mask_rcnn_cbv2_swin_tiny_patch4_window7_mstrain_480-800_adamw_3x_coco.pth'