File size: 5,113 Bytes
9bf4bd7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
157
158
159
160
dictionary = dict(
    type='Dictionary',
    dict_file='{{ fileDirname }}/../../../dicts/lower_english_digits.txt',
    with_start=True,
    with_end=True,
    same_start_end=True,
    with_padding=False,
    with_unknown=False)

model = dict(
    type='ABINet',
    backbone=dict(type='ResNetABI'),
    encoder=dict(
        type='ABIEncoder',
        n_layers=3,
        n_head=8,
        d_model=512,
        d_inner=2048,
        dropout=0.1,
        max_len=8 * 32,
    ),
    decoder=dict(
        type='ABIFuser',
        vision_decoder=dict(
            type='ABIVisionDecoder',
            in_channels=512,
            num_channels=64,
            attn_height=8,
            attn_width=32,
            attn_mode='nearest',
            init_cfg=dict(type='Xavier', layer='Conv2d')),
        module_loss=dict(type='ABIModuleLoss', letter_case='lower'),
        postprocessor=dict(type='AttentionPostprocessor'),
        dictionary=dictionary,
        max_seq_len=26,
    ),
    data_preprocessor=dict(
        type='TextRecogDataPreprocessor',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375]))

train_pipeline = [
    dict(type='LoadImageFromFile', ignore_empty=True, min_size=2),
    dict(type='LoadOCRAnnotations', with_text=True),
    dict(type='Resize', scale=(128, 32)),
    dict(
        type='RandomApply',
        prob=0.5,
        transforms=[
            dict(
                type='RandomChoice',
                transforms=[
                    dict(
                        type='RandomRotate',
                        max_angle=15,
                    ),
                    dict(
                        type='TorchVisionWrapper',
                        op='RandomAffine',
                        degrees=15,
                        translate=(0.3, 0.3),
                        scale=(0.5, 2.),
                        shear=(-45, 45),
                    ),
                    dict(
                        type='TorchVisionWrapper',
                        op='RandomPerspective',
                        distortion_scale=0.5,
                        p=1,
                    ),
                ])
        ],
    ),
    dict(
        type='RandomApply',
        prob=0.25,
        transforms=[
            dict(type='PyramidRescale'),
            dict(
                type='mmdet.Albu',
                transforms=[
                    dict(type='GaussNoise', var_limit=(20, 20), p=0.5),
                    dict(type='MotionBlur', blur_limit=7, p=0.5),
                ]),
        ]),
    dict(
        type='RandomApply',
        prob=0.25,
        transforms=[
            dict(
                type='TorchVisionWrapper',
                op='ColorJitter',
                brightness=0.5,
                saturation=0.5,
                contrast=0.5,
                hue=0.1),
        ]),
    dict(
        type='PackTextRecogInputs',
        meta_keys=('img_path', 'ori_shape', 'img_shape', 'valid_ratio'))
]

test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='Resize', scale=(128, 32)),
    # add loading annotation after ``Resize`` because ground truth
    # does not need to do resize data transform
    dict(type='LoadOCRAnnotations', with_text=True),
    dict(
        type='PackTextRecogInputs',
        meta_keys=('img_path', 'ori_shape', 'img_shape', 'valid_ratio'))
]

tta_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='TestTimeAug',
        transforms=[
            [
                dict(
                    type='ConditionApply',
                    true_transforms=[
                        dict(
                            type='ImgAugWrapper',
                            args=[dict(cls='Rot90', k=0, keep_size=False)])
                    ],
                    condition="results['img_shape'][1]<results['img_shape'][0]"
                ),
                dict(
                    type='ConditionApply',
                    true_transforms=[
                        dict(
                            type='ImgAugWrapper',
                            args=[dict(cls='Rot90', k=1, keep_size=False)])
                    ],
                    condition="results['img_shape'][1]<results['img_shape'][0]"
                ),
                dict(
                    type='ConditionApply',
                    true_transforms=[
                        dict(
                            type='ImgAugWrapper',
                            args=[dict(cls='Rot90', k=3, keep_size=False)])
                    ],
                    condition="results['img_shape'][1]<results['img_shape'][0]"
                ),
            ],
            [dict(type='Resize', scale=(128, 32))],
            # add loading annotation after ``Resize`` because ground truth
            # does not need to do resize data transform
            [dict(type='LoadOCRAnnotations', with_text=True)],
            [
                dict(
                    type='PackTextRecogInputs',
                    meta_keys=('img_path', 'ori_shape', 'img_shape',
                               'valid_ratio'))
            ]
        ])
]