-
Notifications
You must be signed in to change notification settings - Fork 57
/
Copy pathvisionlan_resnet45_LF_2.yaml
173 lines (159 loc) · 4.93 KB
/
visionlan_resnet45_LF_2.yaml
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
161
162
163
164
165
166
167
168
169
170
171
172
173
system:
mode: 0 # 0 for graph mode, 1 for pynative mode in MindSpore
distribute: True
amp_level: 'O0'
seed: 42
log_interval: 200
val_while_train: True
drop_overflow_update: True
common:
character_dict_path: &character_dict_path
num_classes: &num_classes 37 # num_chars_in_dict+1, TODO: retreive it from dict or check correctness
max_text_len: &max_text_len 25
infer_mode: &infer_mode False
use_space_char: &use_space_char False
batch_size: &batch_size 192
blank_at_last: &blank_at_last False # pad blank at the last or the first of the dictionary
lower: &lower True
training_step: &training_step 'LF_2'
model:
type: rec
transform: null
resume: False
backbone:
name: rec_resnet45
pretrained: False
strides: [2, 2, 2, 1, 1]
head:
name: VisionLANHead
n_layers: 3
n_position: 256
n_dim: 512
max_text_length: *max_text_len
training_step: *training_step
pretrained: ./tmp_visionlan/LF_1/e8.ckpt # the initial state dict needs to be provided for LF_2 and LA
postprocess:
name: VisionLANPostProcess
character_dict_path: *character_dict_path
use_space_char: *use_space_char
blank_at_last: *blank_at_last
lower: *lower
max_text_length: *max_text_len
metric:
name: RecMetric
main_indicator: acc
character_dict_path: *character_dict_path
ignore_space: True
print_flag: False
loss:
name: VisionLANLoss
mode: *training_step
weight_res: 0.5
weight_mas: 0.5
scheduler:
scheduler: step_decay
min_lr: 0.0
lr: &init_lr 0.0001
decay_rate: 0.1
decay_epochs: 3
num_epochs: 4
warmup_epochs: 0
optimizer:
opt: adam
weight_decay: 0.0
nesterov: False
# lr: # no need to set lr here because in train.py, the lr scheduler is passed to create_optimizer
grouping_strategy: visionlan
training_step: *training_step
loss_scaler:
type: dynamic
loss_scale: 512
scale_factor: 2.0
scale_window: 1000
train:
ema: True
ema_decay: 0.9999
clip_norm: 20.0
ckpt_save_dir: './tmp_visionlan/LF_2'
dataset_sink_mode: False
pred_cast_fp32: False
dataset:
type: LMDBDataset
dataset_root: ./datasets # Optional, if set, dataset_root will be used as a prefix for data_dir
data_dir: train
# label_file: # not required when using LMDBDataset
sample_ratio: 1.0
shuffle: True
filter_max_len: True
max_text_len: *max_text_len
transform_pipeline:
- DecodeImage:
img_mode: BGR
to_float32: False
- VisionLANLabelEncode:
max_text_len: *max_text_len
character_dict_path: *character_dict_path
use_space_char: *use_space_char
blank_at_last: *blank_at_last
lower: *lower
- SVTRRecAug:
aug_type: 0
deterioration_factor: null
- SVTRRecResizeImg:
image_shape: [64, 256] # H, W
padding: False
- NormalizeImage:
bgr_to_rgb: True
is_hwc: True
mean: [127.0, 127.0, 127.0]
std: [127.0, 127.0, 127.0]
- ToCHWImage:
# the order of the dataloader list, matching the network input and the input labels for the loss function, and optional data for debug/visaulize
output_columns: ['image', 'label', 'label_res', 'label_sub', 'label_id', 'length']
net_input_column_index: [0, 4] # input indices for network forward func in output_columns
label_column_index: [1, 2, 3, 5] # input indices marked as label
loader:
shuffle: True # TODO: tbc
batch_size: *batch_size
drop_remainder: True
max_rowsize: 12
num_workers: 8
eval:
ckpt_load_path: ./tmp_visionlan/LF_2/best.ckpt
dataset_sink_mode: False
dataset:
type: LMDBDataset
dataset_root: ./datasets
data_dir: evaluation/Sumof6benchmarks
# label_file: # not required when using LMDBDataset
sample_ratio: 1.0
shuffle: False
transform_pipeline:
- DecodeImage:
img_mode: BGR
to_float32: False
- VisionLANLabelEncode:
max_text_len: *max_text_len
character_dict_path: *character_dict_path
use_space_char: *use_space_char
blank_at_last: *blank_at_last
lower: *lower
- SVTRRecResizeImg:
image_shape: [64, 256] # H, W
padding: False
- NormalizeImage: # different from paddle (paddle wrongly normalize BGR image with RGB mean/std from ImageNet for det, and simple rescale to [-1, 1] in rec.
bgr_to_rgb: True
is_hwc: True
mean: [127.0, 127.0, 127.0]
std: [127.0, 127.0, 127.0]
- ToCHWImage:
# the order of the dataloader list, matching the network input and the input labels for the loss function, and optional data for debug/visaulize
output_columns: ['image', 'text_padded', 'length']
net_input_column_index: [0] # input indices for network forward func in output_columns
label_column_index: [1, 2]
loader:
shuffle: False # TODO: tbc
batch_size: 64
drop_remainder: False
max_rowsize: 12
num_workers: 8