forked from mindspore-lab/mindocr
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrare_resnet34.yaml
144 lines (132 loc) · 3.83 KB
/
rare_resnet34.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
system:
mode: 0 # 0 for graph mode, 1 for pynative mode in MindSpore
distribute: True
amp_level: "O2"
seed: 42
log_interval: 100
val_while_train: True
drop_overflow_update: False
common:
character_dict_path: &character_dict_path
num_classes: &num_classes 38 # num_chars_in_dict + 2
max_text_len: &max_text_len 25
infer_mode: &infer_mode False
use_space_char: &use_space_char False
batch_size: &batch_size 512
model:
type: rec
transform: null
backbone:
name: rec_resnet34
pretrained: False
neck:
name: RNNEncoder
hidden_size: 256
head:
name: AttentionHead
hidden_size: 256
out_channels: *num_classes
batch_max_length: *max_text_len
postprocess:
name: RecAttnLabelDecode
character_dict_path: *character_dict_path
use_space_char: *use_space_char
metric:
name: RecMetric
main_indicator: acc
character_dict_path: *character_dict_path
ignore_space: True
print_flag: False
loss:
name: AttentionLoss
scheduler:
scheduler: warmup_cosine_decay
min_lr: 0.0
lr: 0.0005
num_epochs: 30
warmup_epochs: 1
decay_epochs: 29
optimizer:
opt: adamw
filter_bias_and_bn: True
weight_decay: 0.05
loss_scaler:
type: dynamic
loss_scale: 512
scale_factor: 2.0
scale_window: 1000
train:
ckpt_save_dir: "./tmp_rec"
dataset_sink_mode: False
dataset:
type: LMDBDataset
dataset_root: path/to/data_lmdb_release/
data_dir: training/
label_file: null
sample_ratio: 1.0
shuffle: True
transform_pipeline:
- DecodeImage:
img_mode: BGR
to_float32: False
- RecAttnLabelEncode:
max_text_len: *max_text_len
character_dict_path: *character_dict_path
use_space_char: *use_space_char
lower: True
- RecResizeImg: # different from paddle (paddle converts image from HWC to CHW and rescale to [-1, 1] after resize.
image_shape: [32, 100] # H, W
infer_mode: *infer_mode
character_dict_path: *character_dict_path
padding: False # aspect ratio will be preserved if true.
- 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:
output_columns: ["image", "text_seq"]
net_input_column_index: [0, 1] # input indices for network forward func in output_columns
label_column_index: [1] # input indices marked as label
loader:
shuffle: True # TODO: tbc
batch_size: *batch_size
drop_remainder: True
max_rowsize: 12
num_workers: 1
eval:
ckpt_load_path: "./tmp_rec/best.ckpt"
dataset_sink_mode: False
dataset:
type: LMDBDataset
dataset_root: path/to/data_lmdb_release/
data_dir: validation/
label_file: null
sample_ratio: 1.0
shuffle: False
transform_pipeline:
- DecodeImage:
img_mode: RGB
to_float32: False
- RecAttnLabelEncode:
max_text_len: *max_text_len
character_dict_path: *character_dict_path
use_space_char: *use_space_char
lower: True
- RecResizeNormForInfer:
target_height: 32
target_width: 100
keep_ratio: False
padding: False
norm_before_pad: False
- 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", "text_length"]
net_input_column_index: [0] # input indices for network forward func in output_columns
label_column_index: [1, 2] # input indices marked as label
loader:
shuffle: False
batch_size: 512
drop_remainder: False
max_rowsize: 12
num_workers: 1