lamonov.pavel/PaddleOCR
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

first commit
Some checks are pending
Build/Publish Develop Docs / deploy (push) Waiting to run
Details

Browse Source
This commit is contained in:
Ламонов Павел
2025-07-02 08:57:16 +03:00
commit 56532cc9a9
1901 changed files with 457695 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
benchmark/PaddleOCR_DBNet/config/SynthText.yaml Normal file
View File

@@ -0,0 +1,40 @@
name: DBNet
dataset:
train:
dataset:
type: SynthTextDataset # 数据集类型
args:
data_path: ''# SynthTextDataset 根目录
pre_processes: # 数据的预处理过程，包含augment和标签制作
- type: IaaAugment # 使用imgaug进行变换
args:
- {'type':Fliplr, 'args':{'p':0.5}}
- {'type': Affine, 'args':{'rotate':[-10,10]}}
- {'type':Resize,'args':{'size':[0.5,3]}}
- type: EastRandomCropData
args:
size: [640,640]
max_tries: 50
keep_ratio: true
- type: MakeBorderMap
args:
shrink_ratio: 0.4
- type: MakeShrinkMap
args:
shrink_ratio: 0.4
min_text_size: 8
transforms: # 对图片进行的变换方式
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
filter_keys: ['img_path','img_name','text_polys','texts','ignore_tags','shape'] # 返回数据之前，从数据字典里删除的key
ignore_tags: ['*', '###']
loader:
batch_size: 1
shuffle: true
num_workers: 0
collate_fn: ''

65
benchmark/PaddleOCR_DBNet/config/SynthText_resnet18_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,65 @@
name: DBNet
base: ['config/SynthText.yaml']
arch:
type: Model
backbone:
type: resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 10
show_images_iter: 50
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path: ./datasets/SynthText
img_mode: RGB
loader:
batch_size: 2
shuffle: true
num_workers: 6
collate_fn: ''

69
benchmark/PaddleOCR_DBNet/config/icdar2015.yaml Normal file
View File

@@ -0,0 +1,69 @@
name: DBNet
dataset:
train:
dataset:
type: ICDAR2015Dataset # 数据集类型
args:
data_path: # 一个存放 img_path \t gt_path的文件
- ''
pre_processes: # 数据的预处理过程，包含augment和标签制作
- type: IaaAugment # 使用imgaug进行变换
args:
- {'type':Fliplr, 'args':{'p':0.5}}
- {'type': Affine, 'args':{'rotate':[-10,10]}}
- {'type':Resize,'args':{'size':[0.5,3]}}
- type: EastRandomCropData
args:
size: [640,640]
max_tries: 50
keep_ratio: true
- type: MakeBorderMap
args:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- type: MakeShrinkMap
args:
shrink_ratio: 0.4
min_text_size: 8
transforms: # 对图片进行的变换方式
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
filter_keys: [img_path,img_name,text_polys,texts,ignore_tags,shape] # 返回数据之前，从数据字典里删除的key
ignore_tags: ['*', '###']
loader:
batch_size: 1
shuffle: true
num_workers: 0
collate_fn: ''
validate:
dataset:
type: ICDAR2015Dataset
args:
data_path:
- ''
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
transforms:
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
filter_keys: []
ignore_tags: ['*', '###']
loader:
batch_size: 1
shuffle: true
num_workers: 0
collate_fn: ICDARCollectFN

82
benchmark/PaddleOCR_DBNet/config/icdar2015_dcn_resnet18_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,82 @@
name: DBNet
base: ['config/icdar2015.yaml']
arch:
type: Model
backbone:
type: deformable_resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 10
show_images_iter: 50
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.txt
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.txt
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

82
benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,82 @@
name: DBNet
base: ['config/icdar2015.yaml']
arch:
type: Model
backbone:
type: resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 10
show_images_iter: 50
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.txt
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.txt
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

83
benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR_finetune.yaml Normal file
View File

@@ -0,0 +1,83 @@
name: DBNet
base: ['config/icdar2015.yaml']
arch:
type: Model
backbone:
type: resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: StepLR
args:
step_size: 10
gama: 0.8
trainer:
seed: 2
epochs: 500
log_iter: 10
show_images_iter: 50
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.txt
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.txt
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

79
benchmark/PaddleOCR_DBNet/config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,79 @@
name: DBNet
base: ['config/icdar2015.yaml']
arch:
type: Model
backbone:
type: resnet50
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
lr_scheduler:
type: Polynomial
args:
learning_rate: 0.001
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 10
show_images_iter: 50
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output/fp16_o2
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.txt
img_mode: RGB
loader:
batch_size: 16
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.txt
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

73
benchmark/PaddleOCR_DBNet/config/open_dataset.yaml Normal file
View File

@@ -0,0 +1,73 @@
name: DBNet
dataset:
train:
dataset:
type: DetDataset # 数据集类型
args:
data_path: # 一个存放 img_path \t gt_path的文件
- ''
pre_processes: # 数据的预处理过程，包含augment和标签制作
- type: IaaAugment # 使用imgaug进行变换
args:
- {'type':Fliplr, 'args':{'p':0.5}}
- {'type': Affine, 'args':{'rotate':[-10,10]}}
- {'type':Resize,'args':{'size':[0.5,3]}}
- type: EastRandomCropData
args:
size: [640,640]
max_tries: 50
keep_ratio: true
- type: MakeBorderMap
args:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- type: MakeShrinkMap
args:
shrink_ratio: 0.4
min_text_size: 8
transforms: # 对图片进行的变换方式
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
load_char_annotation: false
expand_one_char: false
filter_keys: [img_path,img_name,text_polys,texts,ignore_tags,shape] # 返回数据之前，从数据字典里删除的key
ignore_tags: ['*', '###']
loader:
batch_size: 1
shuffle: true
num_workers: 0
collate_fn: ''
validate:
dataset:
type: DetDataset
args:
data_path:
- ''
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
transforms:
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
load_char_annotation: false # 是否加载字符级标注
expand_one_char: false # 是否对只有一个字符的框进行宽度扩充，扩充后w = w+h
filter_keys: []
ignore_tags: ['*', '###']
loader:
batch_size: 1
shuffle: true
num_workers: 0
collate_fn: ICDARCollectFN

86
benchmark/PaddleOCR_DBNet/config/open_dataset_dcn_resnet50_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,86 @@
name: DBNet
base: ['config/open_dataset.yaml']
arch:
type: Model
backbone:
type: deformable_resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 1
show_images_iter: 1
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.json
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 2
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.json
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

86
benchmark/PaddleOCR_DBNet/config/open_dataset_resnest50_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,86 @@
name: DBNet
base: ['config/open_dataset.yaml']
arch:
type: Model
backbone:
type: resnest50
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 1
show_images_iter: 1
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.json
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 2
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.json
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN

93
benchmark/PaddleOCR_DBNet/config/open_dataset_resnet18_FPN_DBhead_polyLR.yaml Normal file
View File

@@ -0,0 +1,93 @@
name: DBNet
base: ['config/open_dataset.yaml']
arch:
type: Model
backbone:
type: resnet18
pretrained: true
neck:
type: FPN
inner_channels: 256
head:
type: DBHead
out_channels: 2
k: 50
post_processing:
type: SegDetectorRepresenter
args:
thresh: 0.3
box_thresh: 0.7
max_candidates: 1000
unclip_ratio: 1.5 # from paper
metric:
type: QuadMetric
args:
is_output_polygon: false
loss:
type: DBLoss
alpha: 1
beta: 10
ohem_ratio: 3
optimizer:
type: Adam
args:
lr: 0.001
weight_decay: 0
amsgrad: true
lr_scheduler:
type: WarmupPolyLR
args:
warmup_epoch: 3
trainer:
seed: 2
epochs: 1200
log_iter: 1
show_images_iter: 1
resume_checkpoint: ''
finetune_checkpoint: ''
output_dir: output
visual_dl: false
amp:
scale_loss: 1024
amp_level: O2
custom_white_list: []
custom_black_list: ['exp', 'sigmoid', 'concat']
dataset:
train:
dataset:
args:
data_path:
- ./datasets/train.json
transforms: # 对图片进行的变换方式
- type: ToTensor
args: {}
- type: Normalize
args:
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 2
shuffle: true
num_workers: 6
collate_fn: ''
validate:
dataset:
args:
data_path:
- ./datasets/test.json
pre_processes:
- type: ResizeShortSize
args:
short_size: 736
resize_text_polys: false
img_mode: RGB
load_char_annotation: false
expand_one_char: false
loader:
batch_size: 1
shuffle: true
num_workers: 6
collate_fn: ICDARCollectFN