Skip to content

GitLab

"git@developer.sourcefind.cn:chenpangpang/transformers.git" did not exist on "da37eb8e43ee48e7e2e5d90869c4905e19fba3a0"
Commit e40fd431 authored by Leif's avatar Leif
Browse files

Merge remote-tracking branch 'origin/dygraph' into dygraph

parents 6e0cbbe1 0da240d0
Hide whitespace changes
Inline Side-by-side
Showing with 728 additions and 153 deletions
README_ch.md
View file @ e40fd431
...@@ -81,7 +81,7 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力 ...@@ -81,7 +81,7 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力
| 模型简介 | 模型名称 |推荐场景 | 检测模型 | 方向分类器 | 识别模型 | | 模型简介 | 模型名称 |推荐场景 | 检测模型 | 方向分类器 | 识别模型 |
| ------------ | --------------- | ----------------|---- | ---------- | -------- | | ------------ | --------------- | ----------------|---- | ---------- | -------- |
| 中英文超轻量PP-OCRv2模型（13.0M） | ch_PP-OCRv2_xx |移动端&服务器端|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/chinese/ch_PP-OCRv2_det_distill_train.tar)| [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar)| | 中英文超轻量PP-OCRv2模型（13.0M） | ch_PP-OCRv2_xx |移动端&服务器端|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar)| [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar)|
| 中英文超轻量PP-OCR mobile模型（9.4M） | ch_ppocr_mobile_v2.0_xx |移动端&服务器端|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar)|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) | | 中英文超轻量PP-OCR mobile模型（9.4M） | ch_ppocr_mobile_v2.0_xx |移动端&服务器端|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar)|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
| 中英文通用PP-OCR server模型（143.4M） |ch_ppocr_server_v2.0_xx|服务器端 |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) | | 中英文通用PP-OCR server模型（143.4M） |ch_ppocr_server_v2.0_xx|服务器端 |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
......
configs/det/det_mv3_pse.yml 0 → 100644
View file @ e40fd431
Global:
use_gpu: true
epoch_num: 600
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/det_mv3_pse/
save_epoch_step: 600
# evaluation is run every 63 iterations
eval_batch_step: [ 0,63 ]
cal_metric_during_train: False
pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_en/img_10.jpg
save_res_path: ./output/det_pse/predicts_pse.txt
Architecture:
model_type: det
algorithm: PSE
Transform: null
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
Neck:
name: FPN
out_channels: 96
Head:
name: PSEHead
hidden_dim: 96
out_channels: 7
Loss:
name: PSELoss
alpha: 0.7
ohem_ratio: 3
kernel_sample_mask: pred
reduction: none
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Step
learning_rate: 0.001
step_size: 200
gamma: 0.1
regularizer:
name: 'L2'
factor: 0.0005
PostProcess:
name: PSEPostProcess
thresh: 0
box_thresh: 0.85
min_area: 16
box_type: box # 'box' or 'poly'
scale: 1
Metric:
name: DetMetric
main_indicator: hmean
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
ratio_list: [ 1.0 ]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- ColorJitter:
brightness: 0.12549019607843137
saturation: 0.5
- IaaAugment:
augmenter_args:
- { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
- { 'type': Fliplr, 'args': { 'p': 0.5 } }
- { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
- MakePseGt:
kernel_num: 7
min_shrink_ratio: 0.4
size: 640
- RandomCropImgMask:
size: [ 640,640 ]
main_key: gt_text
crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
- NormalizeImage:
scale: 1./255.
mean: [ 0.485, 0.456, 0.406 ]
std: [ 0.229, 0.224, 0.225 ]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
loader:
shuffle: True
drop_last: False
batch_size_per_card: 16
num_workers: 8
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
ratio_list: [ 1.0 ]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- DetResizeForTest:
limit_side_len: 736
limit_type: min
- NormalizeImage:
scale: 1./255.
mean: [ 0.485, 0.456, 0.406 ]
std: [ 0.229, 0.224, 0.225 ]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
loader:
shuffle: False
drop_last: False
batch_size_per_card: 1 # must be 1
num_workers: 8
\ No newline at end of file
configs/det/det_r50_vd_pse.yml 0 → 100644
View file @ e40fd431
Global:
use_gpu: true
epoch_num: 600
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/det_r50_vd_pse/
save_epoch_step: 600
# evaluation is run every 125 iterations
eval_batch_step: [ 0,125 ]
cal_metric_during_train: False
pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_en/img_10.jpg
save_res_path: ./output/det_pse/predicts_pse.txt
Architecture:
model_type: det
algorithm: PSE
Transform:
Backbone:
name: ResNet
layers: 50
Neck:
name: FPN
out_channels: 256
Head:
name: PSEHead
hidden_dim: 256
out_channels: 7
Loss:
name: PSELoss
alpha: 0.7
ohem_ratio: 3
kernel_sample_mask: pred
reduction: none
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Step
learning_rate: 0.0001
step_size: 200
gamma: 0.1
regularizer:
name: 'L2'
factor: 0.0005
PostProcess:
name: PSEPostProcess
thresh: 0
box_thresh: 0.85
min_area: 16
box_type: box # 'box' or 'poly'
scale: 1
Metric:
name: DetMetric
main_indicator: hmean
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
ratio_list: [ 1.0 ]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- ColorJitter:
brightness: 0.12549019607843137
saturation: 0.5
- IaaAugment:
augmenter_args:
- { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
- { 'type': Fliplr, 'args': { 'p': 0.5 } }
- { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
- MakePseGt:
kernel_num: 7
min_shrink_ratio: 0.4
size: 640
- RandomCropImgMask:
size: [ 640,640 ]
main_key: gt_text
crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
- NormalizeImage:
scale: 1./255.
mean: [ 0.485, 0.456, 0.406 ]
std: [ 0.229, 0.224, 0.225 ]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
loader:
shuffle: True
drop_last: False
batch_size_per_card: 8
num_workers: 8
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
ratio_list: [ 1.0 ]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- DetResizeForTest:
limit_side_len: 736
limit_type: min
- NormalizeImage:
scale: 1./255.
mean: [ 0.485, 0.456, 0.406 ]
std: [ 0.229, 0.224, 0.225 ]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
loader:
shuffle: False
drop_last: False
batch_size_per_card: 1 # must be 1
num_workers: 8
\ No newline at end of file
configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml
View file @ e40fd431
...@@ -4,7 +4,7 @@ Global: ...@@ -4,7 +4,7 @@ Global:
epoch_num: 800 epoch_num: 800
log_smooth_window: 20 log_smooth_window: 20
print_batch_step: 10 print_batch_step: 10
save_model_dir: ./output/rec_chinese_lite_distillation_v2.1 save_model_dir: ./output/rec_mobile_pp-OCRv2
save_epoch_step: 3 save_epoch_step: 3
eval_batch_step: [0, 2000] eval_batch_step: [0, 2000]
cal_metric_during_train: true cal_metric_during_train: true
...@@ -19,7 +19,7 @@ Global: ...@@ -19,7 +19,7 @@ Global:
infer_mode: false infer_mode: false
use_space_char: true use_space_char: true
distributed: true distributed: true
save_res_path: ./output/rec/predicts_chinese_lite_distillation_v2.1.txt save_res_path: ./output/rec/predicts_mobile_pp-OCRv2.txt
Optimizer: Optimizer:
...@@ -35,79 +35,32 @@ Optimizer: ...@@ -35,79 +35,32 @@ Optimizer:
name: L2 name: L2
factor: 2.0e-05 factor: 2.0e-05
Architecture: Architecture:
model_type: &model_type "rec" model_type: rec
name: DistillationModel algorithm: CRNN
algorithm: Distillation Transform:
Models: Backbone:
Teacher: name: MobileNetV1Enhance
pretrained: scale: 0.5
freeze_params: false Neck:
return_all_feats: true name: SequenceEncoder
model_type: *model_type encoder_type: rnn
algorithm: CRNN hidden_size: 64
Transform: Head:
Backbone: name: CTCHead
name: MobileNetV1Enhance mid_channels: 96
scale: 0.5 fc_decay: 0.00002
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00002
Student:
pretrained:
freeze_params: false
return_all_feats: true
model_type: *model_type
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV1Enhance
scale: 0.5
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00002
Loss: Loss:
name: CombinedLoss name: CTCLoss
loss_config_list:
- DistillationCTCLoss:
weight: 1.0
model_name_list: ["Student", "Teacher"]
key: head_out
- DistillationDMLLoss:
weight: 1.0
act: "softmax"
model_name_pairs:
- ["Student", "Teacher"]
key: head_out
- DistillationDistanceLoss:
weight: 1.0
mode: "l2"
model_name_pairs:
- ["Student", "Teacher"]
key: backbone_out
PostProcess: PostProcess:
name: DistillationCTCLabelDecode name: CTCLabelDecode
model_name: ["Student", "Teacher"]
key: head_out
Metric: Metric:
name: DistillationMetric name: RecMetric
base_metric_name: RecMetric
main_indicator: acc main_indicator: acc
key: "Student"
Train: Train:
dataset: dataset:
...@@ -132,7 +85,6 @@ Train: ...@@ -132,7 +85,6 @@ Train:
shuffle: true shuffle: true
batch_size_per_card: 128 batch_size_per_card: 128
drop_last: true drop_last: true
num_sections: 1
num_workers: 8 num_workers: 8
Eval: Eval:
dataset: dataset:
......
configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml 0 → 100644
View file @ e40fd431
Global:
debug: false
use_gpu: true
epoch_num: 800
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_pp-OCRv2_distillation
save_epoch_step: 3
eval_batch_step: [0, 2000]
cal_metric_during_train: true
pretrained_model:
checkpoints:
save_inference_dir:
use_visualdl: false
infer_img: doc/imgs_words/ch/word_1.jpg
character_dict_path: ppocr/utils/ppocr_keys_v1.txt
character_type: ch
max_text_length: 25
infer_mode: false
use_space_char: true
distributed: true
save_res_path: ./output/rec/predicts_pp-OCRv2_distillation.txt
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Piecewise
decay_epochs : [700, 800]
values : [0.001, 0.0001]
warmup_epoch: 5
regularizer:
name: L2
factor: 2.0e-05
Architecture:
model_type: &model_type "rec"
name: DistillationModel
algorithm: Distillation
Models:
Teacher:
pretrained:
freeze_params: false
return_all_feats: true
model_type: *model_type
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV1Enhance
scale: 0.5
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00002
Student:
pretrained:
freeze_params: false
return_all_feats: true
model_type: *model_type
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV1Enhance
scale: 0.5
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00002
Loss:
name: CombinedLoss
loss_config_list:
- DistillationCTCLoss:
weight: 1.0
model_name_list: ["Student", "Teacher"]
key: head_out
- DistillationDMLLoss:
weight: 1.0
act: "softmax"
use_log: true
model_name_pairs:
- ["Student", "Teacher"]
key: head_out
- DistillationDistanceLoss:
weight: 1.0
mode: "l2"
model_name_pairs:
- ["Student", "Teacher"]
key: backbone_out
PostProcess:
name: DistillationCTCLabelDecode
model_name: ["Student", "Teacher"]
key: head_out
Metric:
name: DistillationMetric
base_metric_name: RecMetric
main_indicator: acc
key: "Student"
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/
label_file_list:
- ./train_data/train_list.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- RecAug:
- CTCLabelEncode:
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys:
- image
- label
- length
loader:
shuffle: true
batch_size_per_card: 128
drop_last: true
num_sections: 1
num_workers: 8
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data
label_file_list:
- ./train_data/val_list.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- CTCLabelEncode:
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys:
- image
- label
- length
loader:
shuffle: false
drop_last: false
batch_size_per_card: 128
num_workers: 8
configs/rec/rec_mtb_nrtr.yml
View file @ e40fd431
...@@ -46,7 +46,7 @@ Architecture: ...@@ -46,7 +46,7 @@ Architecture:
name: Transformer name: Transformer
d_model: 512 d_model: 512
num_encoder_layers: 6 num_encoder_layers: 6
beam_size: 10 # When Beam size is greater than 0, it means to use beam search when evaluation. beam_size: -1 # When Beam size is greater than 0, it means to use beam search when evaluation.
Loss: Loss:
...@@ -65,7 +65,7 @@ Train: ...@@ -65,7 +65,7 @@ Train:
name: LMDBDataSet name: LMDBDataSet
data_dir: ./train_data/data_lmdb_release/training/ data_dir: ./train_data/data_lmdb_release/training/
transforms: transforms:
- NRTRDecodeImage: # load image - DecodeImage: # load image
img_mode: BGR img_mode: BGR
channel_first: False channel_first: False
- NRTRLabelEncode: # Class handling label - NRTRLabelEncode: # Class handling label
...@@ -85,7 +85,7 @@ Eval: ...@@ -85,7 +85,7 @@ Eval:
name: LMDBDataSet name: LMDBDataSet
data_dir: ./train_data/data_lmdb_release/evaluation/ data_dir: ./train_data/data_lmdb_release/evaluation/
transforms: transforms:
- NRTRDecodeImage: # load image - DecodeImage: # load image
img_mode: BGR img_mode: BGR
channel_first: False channel_first: False
- NRTRLabelEncode: # Class handling label - NRTRLabelEncode: # Class handling label
......
configs/rec/rec_r31_sar.yml
View file @ e40fd431
...@@ -79,7 +79,7 @@ Train: ...@@ -79,7 +79,7 @@ Train:
Eval: Eval:
dataset: dataset:
name: LMDBDataSet name: LMDBDataSet
data_dir: ./eval_data/evaluation/ data_dir: ./train_data/data_lmdb_release/evaluation/
transforms: transforms:
- DecodeImage: # load image - DecodeImage: # load image
img_mode: BGR img_mode: BGR
......
deploy/cpp_infer/src/main.cpp
View file @ e40fd431
...@@ -91,7 +91,7 @@ int main_det(std::vector<cv::String> cv_all_img_names) { ...@@ -91,7 +91,7 @@ int main_det(std::vector<cv::String> cv_all_img_names) {
FLAGS_use_tensorrt, FLAGS_precision); FLAGS_use_tensorrt, FLAGS_precision);
for (int i = 0; i < cv_all_img_names.size(); ++i) { for (int i = 0; i < cv_all_img_names.size(); ++i) {
LOG(INFO) << "The predict img: " << cv_all_img_names[i]; // LOG(INFO) << "The predict img: " << cv_all_img_names[i];
cv::Mat srcimg = cv::imread(cv_all_img_names[i], cv::IMREAD_COLOR); cv::Mat srcimg = cv::imread(cv_all_img_names[i], cv::IMREAD_COLOR);
if (!srcimg.data) { if (!srcimg.data) {
...@@ -106,6 +106,16 @@ int main_det(std::vector<cv::String> cv_all_img_names) { ...@@ -106,6 +106,16 @@ int main_det(std::vector<cv::String> cv_all_img_names) {
time_info[0] += det_times[0]; time_info[0] += det_times[0];
time_info[1] += det_times[1]; time_info[1] += det_times[1];
time_info[2] += det_times[2]; time_info[2] += det_times[2];
if (FLAGS_benchmark) {
cout << cv_all_img_names[i] << '\t';
for (int n = 0; n < boxes.size(); n++) {
for (int m = 0; m < boxes[n].size(); m++) {
cout << boxes[n][m][0] << ' ' << boxes[n][m][1] << ' ';
}
}
cout << endl;
}
} }
if (FLAGS_benchmark) { if (FLAGS_benchmark) {
......
deploy/hubserving/ocr_det/params.py
View file @ e40fd431
...@@ -13,7 +13,7 @@ def read_params(): ...@@ -13,7 +13,7 @@ def read_params():
#params for text detector #params for text detector
cfg.det_algorithm = "DB" cfg.det_algorithm = "DB"
cfg.det_model_dir = "./inference/ch_ppocr_mobile_v2.0_det_infer/" cfg.det_model_dir = "./inference/ch_PP-OCRv2_det_infer/"
cfg.det_limit_side_len = 960 cfg.det_limit_side_len = 960
cfg.det_limit_type = 'max' cfg.det_limit_type = 'max'
......
deploy/hubserving/ocr_rec/params.py
View file @ e40fd431
...@@ -13,7 +13,7 @@ def read_params(): ...@@ -13,7 +13,7 @@ def read_params():
#params for text recognizer #params for text recognizer
cfg.rec_algorithm = "CRNN" cfg.rec_algorithm = "CRNN"
cfg.rec_model_dir = "./inference/ch_ppocr_mobile_v2.0_rec_infer/" cfg.rec_model_dir = "./inference/ch_PP-OCRv2_rec_infer/"
cfg.rec_image_shape = "3, 32, 320" cfg.rec_image_shape = "3, 32, 320"
cfg.rec_char_type = 'ch' cfg.rec_char_type = 'ch'
......
deploy/hubserving/ocr_system/params.py
View file @ e40fd431
...@@ -13,7 +13,7 @@ def read_params(): ...@@ -13,7 +13,7 @@ def read_params():
#params for text detector #params for text detector
cfg.det_algorithm = "DB" cfg.det_algorithm = "DB"
cfg.det_model_dir = "./inference/ch_ppocr_mobile_v2.0_det_infer/" cfg.det_model_dir = "./inference/ch_PP-OCRv2_det_infer/"
cfg.det_limit_side_len = 960 cfg.det_limit_side_len = 960
cfg.det_limit_type = 'max' cfg.det_limit_type = 'max'
...@@ -31,7 +31,7 @@ def read_params(): ...@@ -31,7 +31,7 @@ def read_params():
#params for text recognizer #params for text recognizer
cfg.rec_algorithm = "CRNN" cfg.rec_algorithm = "CRNN"
cfg.rec_model_dir = "./inference/ch_ppocr_mobile_v2.0_rec_infer/" cfg.rec_model_dir = "./inference/ch_PP-OCRv2_rec_infer/"
cfg.rec_image_shape = "3, 32, 320" cfg.rec_image_shape = "3, 32, 320"
cfg.rec_char_type = 'ch' cfg.rec_char_type = 'ch'
......
deploy/hubserving/readme.md
View file @ e40fd431
...@@ -34,10 +34,10 @@ pip3 install paddlehub==2.1.0 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/sim ...@@ -34,10 +34,10 @@ pip3 install paddlehub==2.1.0 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/sim
``` ```
### 2. 下载推理模型 ### 2. 下载推理模型
安装服务模块前，需要准备推理模型并放到正确路径。默认使用的是v2.0版的超轻量模型，默认模型路径为： 安装服务模块前，需要准备推理模型并放到正确路径。默认使用的是PP-OCRv2模型，默认模型路径为：
``` ```
检测模型：./inference/ch_ppocr_mobile_v2.0_det_infer/ 检测模型：./inference/ch_PP-OCRv2_det_infer/
识别模型：./inference/ch_ppocr_mobile_v2.0_rec_infer/ 识别模型：./inference/ch_PP-OCRv2_rec_infer/
方向分类器：./inference/ch_ppocr_mobile_v2.0_cls_infer/ 方向分类器：./inference/ch_ppocr_mobile_v2.0_cls_infer/
``` ```
......
deploy/hubserving/readme_en.md
View file @ e40fd431
...@@ -35,10 +35,10 @@ pip3 install paddlehub==2.1.0 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/sim ...@@ -35,10 +35,10 @@ pip3 install paddlehub==2.1.0 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/sim
``` ```
### 2. Download inference model ### 2. Download inference model
Before installing the service module, you need to prepare the inference model and put it in the correct path. By default, the ultra lightweight model of v2.0 is used, and the default model path is: Before installing the service module, you need to prepare the inference model and put it in the correct path. By default, the PP-OCRv2 models are used, and the default model path is:
``` ```
detection model: ./inference/ch_ppocr_mobile_v2.0_det_infer/ detection model: ./inference/ch_PP-OCRv2_det_infer/
recognition model: ./inference/ch_ppocr_mobile_v2.0_rec_infer/ recognition model: ./inference/ch_PP-OCRv2_rec_infer/
text direction classifier: ./inference/ch_ppocr_mobile_v2.0_cls_infer/ text direction classifier: ./inference/ch_ppocr_mobile_v2.0_cls_infer/
``` ```
......
deploy/slim/quantization/quant_kl.py 0 → 100755
View file @ e40fd431
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '..', '..', '..')))
sys.path.append(
os.path.abspath(os.path.join(__dir__, '..', '..', '..', 'tools')))
import yaml
import paddle
import paddle.distributed as dist
paddle.seed(2)
from ppocr.data import build_dataloader
from ppocr.modeling.architectures import build_model
from ppocr.losses import build_loss
from ppocr.optimizer import build_optimizer
from ppocr.postprocess import build_post_process
from ppocr.metrics import build_metric
from ppocr.utils.save_load import init_model
import tools.program as program
import paddleslim
from paddleslim.dygraph.quant import QAT
import numpy as np
dist.get_world_size()
class PACT(paddle.nn.Layer):
def __init__(self):
super(PACT, self).__init__()
alpha_attr = paddle.ParamAttr(
name=self.full_name() + ".pact",
initializer=paddle.nn.initializer.Constant(value=20),
learning_rate=1.0,
regularizer=paddle.regularizer.L2Decay(2e-5))
self.alpha = self.create_parameter(
shape=[1], attr=alpha_attr, dtype='float32')
def forward(self, x):
out_left = paddle.nn.functional.relu(x - self.alpha)
out_right = paddle.nn.functional.relu(-self.alpha - x)
x = x - out_left + out_right
return x
quant_config = {
# weight preprocess type, default is None and no preprocessing is performed.
'weight_preprocess_type': None,
# activation preprocess type, default is None and no preprocessing is performed.
'activation_preprocess_type': None,
# weight quantize type, default is 'channel_wise_abs_max'
'weight_quantize_type': 'channel_wise_abs_max',
# activation quantize type, default is 'moving_average_abs_max'
'activation_quantize_type': 'moving_average_abs_max',
# weight quantize bit num, default is 8
'weight_bits': 8,
# activation quantize bit num, default is 8
'activation_bits': 8,
# data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
'dtype': 'int8',
# window size for 'range_abs_max' quantization. default is 10000
'window_size': 10000,
# The decay coefficient of moving average, default is 0.9
'moving_rate': 0.9,
# for dygraph quantization, layers of type in quantizable_layer_type will be quantized
'quantizable_layer_type': ['Conv2D', 'Linear'],
}
def sample_generator(loader):
def __reader__():
for indx, data in enumerate(loader):
images = np.array(data[0])
yield images
return __reader__
def main(config, device, logger, vdl_writer):
# init dist environment
if config['Global']['distributed']:
dist.init_parallel_env()
global_config = config['Global']
# build dataloader
config['Train']['loader']['num_workers'] = 0
train_dataloader = build_dataloader(config, 'Train', device, logger)
if config['Eval']:
config['Eval']['loader']['num_workers'] = 0
valid_dataloader = build_dataloader(config, 'Eval', device, logger)
else:
valid_dataloader = None
paddle.enable_static()
place = paddle.CPUPlace()
exe = paddle.static.Executor(place)
if 'inference_model' in global_config.keys(): # , 'inference_model'):
inference_model_dir = global_config['inference_model']
else:
inference_model_dir = os.path.dirname(global_config['pretrained_model'])
if not (os.path.exists(os.path.join(inference_model_dir, "inference.pdmodel")) and \
os.path.exists(os.path.join(inference_model_dir, "inference.pdiparams")) ):
raise ValueError(
"Please set inference model dir in Global.inference_model or Global.pretrained_model for post-quantazition"
)
paddleslim.quant.quant_post_static(
executor=exe,
model_dir=inference_model_dir,
model_filename='inference.pdmodel',
params_filename='inference.pdiparams',
quantize_model_path=global_config['save_inference_dir'],
sample_generator=sample_generator(train_dataloader),
save_model_filename='inference.pdmodel',
save_params_filename='inference.pdiparams',
batch_size=1,
batch_nums=None)
if __name__ == '__main__':
config, device, logger, vdl_writer = program.preprocess(is_train=True)
main(config, device, logger, vdl_writer)
doc/doc_ch/algorithm_overview.md
View file @ e40fd431
...@@ -9,11 +9,13 @@ ...@@ -9,11 +9,13 @@
### 1.文本检测算法 ### 1.文本检测算法
PaddleOCR开源的文本检测算法列表： PaddleOCR开源的文本检测算法列表：
- [x] DB([paper]( https://arxiv.org/abs/1911.08947)) [2]（ppocr推荐） - [x] DB([paper]( https://arxiv.org/abs/1911.08947))（ppocr推荐）
- [x] EAST([paper](https://arxiv.org/abs/1704.03155))[1] - [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] - [x] SAST([paper](https://arxiv.org/abs/1908.05498))
- [x] PSENet([paper](https://arxiv.org/abs/1903.12473v2)）
在ICDAR2015文本检测公开数据集上，算法效果如下： 在ICDAR2015文本检测公开数据集上，算法效果如下：
|模型|骨干网络|precision|recall|Hmean|下载链接| |模型|骨干网络|precision|recall|Hmean|下载链接|
| --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- |
|EAST|ResNet50_vd|85.80%|86.71%|86.25%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)| |EAST|ResNet50_vd|85.80%|86.71%|86.25%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
...@@ -21,6 +23,8 @@ PaddleOCR开源的文本检测算法列表： ...@@ -21,6 +23,8 @@ PaddleOCR开源的文本检测算法列表：
|DB|ResNet50_vd|86.41%|78.72%|82.38%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|ResNet50_vd|86.41%|78.72%|82.38%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|77.29%|73.08%|75.12%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |DB|MobileNetV3|77.29%|73.08%|75.12%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
|SAST|ResNet50_vd|91.39%|83.77%|87.42%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)| |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
|PSE|ResNet50_vd|85.81%|79.53%|82.55%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
|PSE|MobileNetV3|82.20%|70.48%|75.89%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
在Total-text文本检测公开数据集上，算法效果如下： 在Total-text文本检测公开数据集上，算法效果如下：
...@@ -39,15 +43,15 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训 ...@@ -39,15 +43,15 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训
### 2.文本识别算法 ### 2.文本识别算法
PaddleOCR基于动态图开源的文本识别算法列表： PaddleOCR基于动态图开源的文本识别算法列表：
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7]（ppocr推荐） - [x] CRNN([paper](https://arxiv.org/abs/1507.05717))（ppocr推荐）
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] - [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] - [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] - [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))
- [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] - [x] SRN([paper](https://arxiv.org/abs/2003.12294))
- [x] NRTR([paper](https://arxiv.org/abs/1806.00926v2)) - [x] NRTR([paper](https://arxiv.org/abs/1806.00926v2))
- [x] SAR([paper](https://arxiv.org/abs/1811.00751v2)) - [x] SAR([paper](https://arxiv.org/abs/1811.00751v2))
参考[DTRB][3](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下： 参考[DTRB](https://arxiv.org/abs/1904.01906) 文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
|模型|骨干网络|Avg Accuracy|模型存储命名|下载链接| |模型|骨干网络|Avg Accuracy|模型存储命名|下载链接|
|---|---|---|---|---| |---|---|---|---|---|
......
doc/doc_ch/benchmark.md
View file @ e40fd431
...@@ -12,40 +12,27 @@ ...@@ -12,40 +12,27 @@
## 评估指标 ## 评估指标
说明： 说明：
- v1.0是未添加优化策略的DB+CRNN模型，v1.1是添加多种优化策略和方向分类器的PP-OCR模型。slim_v1.1是使用裁剪或量化的模型。
- 检测输入图像的的长边尺寸是960。 - 检测输入图像的的长边尺寸是960。
- 评估耗时阶段为图像输入到结果输出的完整阶段，包括了图像的预处理和后处理。 - 评估耗时阶段为图像预测耗时，不包括图像的预处理和后处理。
- `Intel至强6148`为服务器端CPU型号，测试中使用Intel MKL-DNN 加速。 - `Intel至强6148`为服务器端CPU型号，测试中使用Intel MKL-DNN 加速。
- `骁龙855`为移动端处理平台型号。 - `骁龙855`为移动端处理平台型号。
不同预测模型大小和整体识别精度对比 预测模型大小和整体识别精度对比
| 模型名称 | 整体模型<br>大小\(M\) | 检测模型<br>大小\(M\) | 方向分类器<br>模型大小\(M\) | 识别模型<br>大小\(M\) | 整体识别<br>F\-score | | 模型名称 | 整体模型<br>大小\(M\) | 检测模型<br>大小\(M\) | 方向分类器<br>模型大小\(M\) | 识别模型<br>大小\(M\) | 整体识别<br>F\-score |
|:-:|:-:|:-:|:-:|:-:|:-:| |:-:|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.5193 | | PP-OCRv2 | 11\.6 | 3\.0 | 0\.9 | 8\.6 | 0\.5224 |
| ch\_ppocr\_server\_v1\.1 | 155\.1 | 47\.2 | 0\.9 | 107 | 0\.5414 | | PP-OCR mobile | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.503 |
| ch\_ppocr\_mobile\_v1\.0 | 8\.6 | 4\.1 | \- | 4\.5 | 0\.393 | | PP-OCR server | 155\.1 | 47\.2 | 0\.9 | 107 | 0\.570 |
| ch\_ppocr\_server\_v1\.0 | 203\.8 | 98\.5 | \- | 105\.3 | 0\.4436 |
不同预测模型在T4 GPU上预测速度对比，单位ms
| 模型名称 | 整体 | 检测 | 方向分类器 | 识别 |
|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 137 | 35 | 24 | 78 |
| ch\_ppocr\_server\_v1\.1 | 204 | 39 | 25 | 140 |
| ch\_ppocr\_mobile\_v1\.0 | 117 | 41 | \- | 76 |
| ch\_ppocr\_server\_v1\.0 | 199 | 52 | \- | 147 |
不同预测模型在CPU上预测速度对比，单位ms
| 模型名称 | 整体 | 检测 | 方向分类器 | 识别 | 预测模型在CPU和GPU上的速度对比，单位ms
|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 421 | 164 | 51 | 206 |
| ch\_ppocr\_mobile\_v1\.0 | 398 | 219 | \- | 179 |
裁剪量化模型和原始模型模型大小，整体识别精度和在SD 855上预测速度对比 | 模型名称 | CPU | T4 GPU |
|:-:|:-:|:-:|
| PP-OCRv2 | 330 | 111 |
| PP-OCR mobile | 356 | 11 6|
| PP-OCR server | 1056 | 200 |
| 模型名称 | 整体模型<br>大小\(M\) | 检测模型<br>大小\(M\) | 方向分类器<br>模型大小\(M\) | 识别模型<br>大小\(M\) | 整体识别<br>F\-score | SD 855<br>\(ms\) | 更多 PP-OCR 系列模型的预测指标可以参考[PP-OCR Benchamrk](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/doc/doc_ch/benchmark.md)
|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.5193 | 306 |
| ch\_ppocr\_mobile\_slim\_v1\.1 | 3\.5 | 1\.4 | 0\.5 | 1\.6 | 0\.521 | 268 |
doc/doc_ch/knowledge_distillation.md
View file @ e40fd431
...@@ -39,7 +39,7 @@ PaddleOCR中集成了知识蒸馏的算法，具体地，有以下几个主要 ...@@ -39,7 +39,7 @@ PaddleOCR中集成了知识蒸馏的算法，具体地，有以下几个主要
### 2.1 识别配置文件解析 ### 2.1 识别配置文件解析
配置文件在[ch_PP-OCRv2_rec.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml)。 配置文件在[ch_PP-OCRv2_rec_distillation.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml)。
#### 2.1.1 模型结构 #### 2.1.1 模型结构
...@@ -246,6 +246,39 @@ Metric: ...@@ -246,6 +246,39 @@ Metric:
关于`DistillationMetric`更加具体的实现可以参考: [distillation_metric.py](../../ppocr/metrics/distillation_metric.py#L24)。 关于`DistillationMetric`更加具体的实现可以参考: [distillation_metric.py](../../ppocr/metrics/distillation_metric.py#L24)。
#### 2.1.5 蒸馏模型微调
对蒸馏得到的识别蒸馏进行微调有2种方式。
（1）基于知识蒸馏的微调：这种情况比较简单，下载预训练模型，在[ch_PP-OCRv2_rec_distillation.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml)中配置好预训练模型路径以及自己的数据路径，即可进行模型微调训练。
（2）微调时不使用知识蒸馏：这种情况，需要首先将预训练模型中的学生模型参数提取出来，具体步骤如下。
* 首先下载预训练模型并解压。
```shell
# 下面预训练模型并解压
wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar
tar -xf ch_PP-OCRv2_rec_train.tar
```
* 然后使用python，对其中的学生模型参数进行提取
```python
import paddle
# 加载预训练模型
all_params = paddle.load("ch_PP-OCRv2_rec_train/best_accuracy.pdparams")
# 查看权重参数的keys
print(all_params.keys())
# 学生模型的权重提取
s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
# 查看学生模型权重参数的keys
print(s_params.keys())
# 保存
paddle.save(s_params, "ch_PP-OCRv2_rec_train/student.pdparams")
```
转化完成之后，使用[ch_PP-OCRv2_rec.yml](../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml)，修改预训练模型的路径（为导出的`student.pdparams`模型路径）以及自己的数据路径，即可进行模型微调。
### 2.2 检测配置文件解析 ### 2.2 检测配置文件解析
* coming soon! * coming soon!
doc/doc_en/algorithm_overview_en.md
View file @ e40fd431
...@@ -11,9 +11,10 @@ This tutorial lists the text detection algorithms and text recognition algorithm ...@@ -11,9 +11,10 @@ This tutorial lists the text detection algorithms and text recognition algorithm
### 1. Text Detection Algorithm ### 1. Text Detection Algorithm
PaddleOCR open source text detection algorithms list: PaddleOCR open source text detection algorithms list:
- [x] EAST([paper](https://arxiv.org/abs/1704.03155))[2] - [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] DB([paper](https://arxiv.org/abs/1911.08947))[1] - [x] DB([paper](https://arxiv.org/abs/1911.08947))
- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] - [x] SAST([paper](https://arxiv.org/abs/1908.05498))
- [x] PSE([paper](https://arxiv.org/abs/1903.12473v2))
On the ICDAR2015 dataset, the text detection result is as follows: On the ICDAR2015 dataset, the text detection result is as follows:
...@@ -24,6 +25,8 @@ On the ICDAR2015 dataset, the text detection result is as follows: ...@@ -24,6 +25,8 @@ On the ICDAR2015 dataset, the text detection result is as follows:
|DB|ResNet50_vd|86.41%|78.72%|82.38%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|ResNet50_vd|86.41%|78.72%|82.38%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|77.29%|73.08%|75.12%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |DB|MobileNetV3|77.29%|73.08%|75.12%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
|SAST|ResNet50_vd|91.39%|83.77%|87.42%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)| |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
|PSE|ResNet50_vd|85.81%|79.53%|82.55%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
|PSE|MobileNetV3|82.20%|70.48%|75.89%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
On Total-Text dataset, the text detection result is as follows: On Total-Text dataset, the text detection result is as follows:
...@@ -41,11 +44,11 @@ For the training guide and use of PaddleOCR text detection algorithms, please re ...@@ -41,11 +44,11 @@ For the training guide and use of PaddleOCR text detection algorithms, please re
### 2. Text Recognition Algorithm ### 2. Text Recognition Algorithm
PaddleOCR open-source text recognition algorithms list: PaddleOCR open-source text recognition algorithms list:
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7] - [x] CRNN([paper](https://arxiv.org/abs/1507.05717))
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] - [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] - [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] - [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))
- [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] - [x] SRN([paper](https://arxiv.org/abs/2003.12294))
- [x] NRTR([paper](https://arxiv.org/abs/1806.00926v2)) - [x] NRTR([paper](https://arxiv.org/abs/1806.00926v2))
- [x] SAR([paper](https://arxiv.org/abs/1811.00751v2)) - [x] SAR([paper](https://arxiv.org/abs/1811.00751v2))
......
doc/doc_en/benchmark_en.md
View file @ e40fd431
...@@ -13,7 +13,6 @@ We collected 300 images for different real application scenarios to evaluate the ...@@ -13,7 +13,6 @@ We collected 300 images for different real application scenarios to evaluate the
## MEASUREMENT ## MEASUREMENT
Explanation: Explanation:
- v1.0 indicates DB+CRNN models without the strategies. v1.1 indicates the PP-OCR models with the strategies and the direction classify. slim_v1.1 indicates the PP-OCR models with prunner or quantization.
- The long size of the input for the text detector is 960. - The long size of the input for the text detector is 960.
...@@ -27,30 +26,16 @@ Compares the model size and F-score: ...@@ -27,30 +26,16 @@ Compares the model size and F-score:
| Model Name | Model Size <br> of the <br> Whole System\(M\) | Model Size <br>of the Text <br> Detector\(M\) | Model Size <br> of the Direction <br> Classifier\(M\) | Model Size<br>of the Text <br> Recognizer \(M\) | F\-score | | Model Name | Model Size <br> of the <br> Whole System\(M\) | Model Size <br>of the Text <br> Detector\(M\) | Model Size <br> of the Direction <br> Classifier\(M\) | Model Size<br>of the Text <br> Recognizer \(M\) | F\-score |
|:-:|:-:|:-:|:-:|:-:|:-:| |:-:|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.5193 | | PP-OCRv2 | 11\.6 | 3\.0 | 0\.9 | 8\.6 | 0\.5224 |
| ch\_ppocr\_server\_v1\.1 | 155\.1 | 47\.2 | 0\.9 | 107 | 0\.5414 | | PP-OCR mobile | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.503 |
| ch\_ppocr\_mobile\_v1\.0 | 8\.6 | 4\.1 | \- | 4\.5 | 0\.393 | | PP-OCR server | 155\.1 | 47\.2 | 0\.9 | 107 | 0\.570 |
| ch\_ppocr\_server\_v1\.0 | 203\.8 | 98\.5 | \- | 105\.3 | 0\.4436 |
Compares the time-consuming on T4 GPU (ms): Compares the time-consuming on CPU and T4 GPU (ms):
| Model Name | Overall | Text Detector | Direction Classifier | Text Recognizer | | Model Name | CPU | T4 GPU |
|:-:|:-:|:-:|:-:|:-:| |:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 137 | 35 | 24 | 78 | | PP-OCRv2 | 330 | 111 |
| ch\_ppocr\_server\_v1\.1 | 204 | 39 | 25 | 140 | | PP-OCR mobile | 356 | 116|
| ch\_ppocr\_mobile\_v1\.0 | 117 | 41 | \- | 76 | | PP-OCR server | 1056 | 200 |
| ch\_ppocr\_server\_v1\.0 | 199 | 52 | \- | 147 |
Compares the time-consuming on CPU (ms): More indicators of PP-OCR series models can be referred to [PP-OCR Benchamrk](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/doc/doc_en/benchmark_en.md)
| Model Name | Overall | Text Detector | Direction Classifier | Text Recognizer |
|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 421 | 164 | 51 | 206 |
| ch\_ppocr\_mobile\_v1\.0 | 398 | 219 | \- | 179 |
Compares the model size, F-score, the time-consuming on SD 855 of between the slim models and the original models:
| Model Name | Model Size <br> of the <br> Whole System\(M\) | Model Size <br>of the Text <br> Detector\(M\) | Model Size <br> of the Direction <br> Classifier\(M\) | Model Size<br>of the Text <br> Recognizer \(M\) | F\-score | SD 855<br>\(ms\) |
|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
| ch\_ppocr\_mobile\_v1\.1 | 8\.1 | 2\.6 | 0\.9 | 4\.6 | 0\.5193 | 306 |
| ch\_ppocr\_mobile\_slim\_v1\.1 | 3\.5 | 1\.4 | 0\.5 | 1\.6 | 0\.521 | 268 |
ppocr/data/imaug/ColorJitter.py 0 → 100644
View file @ e40fd431
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from paddle.vision.transforms import ColorJitter as pp_ColorJitter
__all__ = ['ColorJitter']
class ColorJitter(object):
def __init__(self, brightness=0, contrast=0, saturation=0, hue=0,**kwargs):
self.aug = pp_ColorJitter(brightness, contrast, saturation, hue)
def __call__(self, data):
image = data['image']
image = self.aug(image)
data['image'] = image
return data
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment