File size: 5,433 Bytes
e8f2571
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
_base_ = [
    '../_base_/datasets/coco_detection.py', '../_base_/default_runtime.py'
]
model = dict(
    type='DETR',
    num_queries=100,
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        bgr_to_rgb=True,
        pad_size_divisor=1),
    backbone=dict(
        type='ResNet',
        depth=50,
        num_stages=4,
        out_indices=(3, ),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=False),
        norm_eval=True,
        style='pytorch',
        init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
    neck=dict(
        type='ChannelMapper',
        in_channels=[2048],
        kernel_size=1,
        out_channels=256,
        act_cfg=None,
        norm_cfg=None,
        num_outs=1),
    encoder=dict(  # DetrTransformerEncoder
        num_layers=6,
        layer_cfg=dict(  # DetrTransformerEncoderLayer
            self_attn_cfg=dict(  # MultiheadAttention
                embed_dims=256,
                num_heads=8,
                dropout=0.1,
                batch_first=True),
            ffn_cfg=dict(
                embed_dims=256,
                feedforward_channels=2048,
                num_fcs=2,
                ffn_drop=0.1,
                act_cfg=dict(type='ReLU', inplace=True)))),
    decoder=dict(  # DetrTransformerDecoder
        num_layers=6,
        layer_cfg=dict(  # DetrTransformerDecoderLayer
            self_attn_cfg=dict(  # MultiheadAttention
                embed_dims=256,
                num_heads=8,
                dropout=0.1,
                batch_first=True),
            cross_attn_cfg=dict(  # MultiheadAttention
                embed_dims=256,
                num_heads=8,
                dropout=0.1,
                batch_first=True),
            ffn_cfg=dict(
                embed_dims=256,
                feedforward_channels=2048,
                num_fcs=2,
                ffn_drop=0.1,
                act_cfg=dict(type='ReLU', inplace=True))),
        return_intermediate=True),
    positional_encoding=dict(num_feats=128, normalize=True),
    bbox_head=dict(
        type='DETRHead',
        num_classes=80,
        embed_dims=256,
        loss_cls=dict(
            type='CrossEntropyLoss',
            bg_cls_weight=0.1,
            use_sigmoid=False,
            loss_weight=1.0,
            class_weight=1.0),
        loss_bbox=dict(type='L1Loss', loss_weight=5.0),
        loss_iou=dict(type='GIoULoss', loss_weight=2.0)),
    # training and testing settings
    train_cfg=dict(
        assigner=dict(
            type='HungarianAssigner',
            match_costs=[
                dict(type='ClassificationCost', weight=1.),
                dict(type='BBoxL1Cost', weight=5.0, box_format='xywh'),
                dict(type='IoUCost', iou_mode='giou', weight=2.0)
            ])),
    test_cfg=dict(max_per_img=100))

# train_pipeline, NOTE the img_scale and the Pad's size_divisor is different
# from the default setting in mmdet.
train_pipeline = [
    dict(type='LoadImageFromFile', backend_args={{_base_.backend_args}}),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='RandomFlip', prob=0.5),
    dict(
        type='RandomChoice',
        transforms=[[
            dict(
                type='RandomChoiceResize',
                scales=[(480, 1333), (512, 1333), (544, 1333), (576, 1333),
                        (608, 1333), (640, 1333), (672, 1333), (704, 1333),
                        (736, 1333), (768, 1333), (800, 1333)],
                keep_ratio=True)
        ],
                    [
                        dict(
                            type='RandomChoiceResize',
                            scales=[(400, 1333), (500, 1333), (600, 1333)],
                            keep_ratio=True),
                        dict(
                            type='RandomCrop',
                            crop_type='absolute_range',
                            crop_size=(384, 600),
                            allow_negative_crop=True),
                        dict(
                            type='RandomChoiceResize',
                            scales=[(480, 1333), (512, 1333), (544, 1333),
                                    (576, 1333), (608, 1333), (640, 1333),
                                    (672, 1333), (704, 1333), (736, 1333),
                                    (768, 1333), (800, 1333)],
                            keep_ratio=True)
                    ]]),
    dict(type='PackDetInputs')
]
train_dataloader = dict(dataset=dict(pipeline=train_pipeline))

# optimizer
optim_wrapper = dict(
    type='OptimWrapper',
    optimizer=dict(type='AdamW', lr=0.0001, weight_decay=0.0001),
    clip_grad=dict(max_norm=0.1, norm_type=2),
    paramwise_cfg=dict(
        custom_keys={'backbone': dict(lr_mult=0.1, decay_mult=1.0)}))

# learning policy
max_epochs = 150
train_cfg = dict(
    type='EpochBasedTrainLoop', max_epochs=max_epochs, val_interval=1)
val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop')

param_scheduler = [
    dict(
        type='MultiStepLR',
        begin=0,
        end=max_epochs,
        by_epoch=True,
        milestones=[100],
        gamma=0.1)
]

# NOTE: `auto_scale_lr` is for automatically scaling LR,
# USER SHOULD NOT CHANGE ITS VALUES.
# base_batch_size = (8 GPUs) x (2 samples per GPU)
auto_scale_lr = dict(base_batch_size=16)