diff --git a/configs/det/det_mv3_pse.yml b/configs/det/det_mv3_pse.yml new file mode 100644 index 0000000000000000000000000000000000000000..61ac24727acbd4f0b1eea15af08c0f9e71ce95a3 --- /dev/null +++ b/configs/det/det_mv3_pse.yml @@ -0,0 +1,135 @@ +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 diff --git a/configs/det/det_r50_vd_pse.yml b/configs/det/det_r50_vd_pse.yml index 63c43d48d3dbcef0ec6c3c360e5aecc5a794180d..4629210747d3b61344cc47b11dcff01e6b738586 100644 --- a/configs/det/det_r50_vd_pse.yml +++ b/configs/det/det_r50_vd_pse.yml @@ -4,16 +4,16 @@ Global: log_smooth_window: 20 print_batch_step: 10 save_model_dir: ./output/det_r50_vd_pse/ - save_epoch_step: 1200 - # evaluation is run every 2000 iterations - eval_batch_step: [0,125] + save_epoch_step: 600 + # evaluation is run every 125 iterations + eval_batch_step: [ 0,125 ] cal_metric_during_train: False - pretrained_model: /ssd1/zhoujun20/fuxian/ResNet50_vd_ssld_pretrained + 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_db/predicts_db.txt + save_res_path: ./output/det_pse/predicts_pse.txt Architecture: model_type: det @@ -68,7 +68,7 @@ Train: data_dir: ./train_data/icdar2015/text_localization/ label_file_list: - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt - ratio_list: [1.0] + ratio_list: [ 1.0 ] transforms: - DecodeImage: # load image img_mode: BGR @@ -81,23 +81,23 @@ Train: augmenter_args: - { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } } - { 'type': Fliplr, 'args': { 'p': 0.5 } } - - { 'type': Affine, 'args': { 'rotate': [-10, 10] } } + - { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } } - MakePseGt: kernel_num: 7 min_shrink_ratio: 0.4 size: 640 - RandomCropImgMask: - size: [640,640] + size: [ 640,640 ] main_key: gt_text - crop_keys: ['image', 'gt_text', 'gt_kernels', 'mask'] + 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] + 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 + keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list loader: shuffle: True drop_last: False @@ -119,15 +119,14 @@ Eval: - DetResizeForTest: limit_side_len: 736 limit_type: min - # resize_long: 2240 - NormalizeImage: scale: 1./255. - mean: [0.485, 0.456, 0.406] - std: [0.229, 0.224, 0.225] + 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'] + keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ] loader: shuffle: False drop_last: False diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index 19d7a69c7fb08a8e7fb36c3043aa211de19b9295..9ba50ffe3c6f94361599ec3835785679c949aa2a 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -9,11 +9,13 @@ ### 1.文本检测算法 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] EAST([paper](https://arxiv.org/abs/1704.03155))[1] -- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] +- [x] DB([paper]( https://arxiv.org/abs/1911.08947))（ppocr推荐） +- [x] EAST([paper](https://arxiv.org/abs/1704.03155)) +- [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： + |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)| @@ -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|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)| +|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%|70.47%|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： @@ -39,13 +43,13 @@ 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 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] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] -- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] -- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] -- [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] +- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))（ppocr推荐） +- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085)) +- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) +- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1)) +- [x] SRN([paper](https://arxiv.org/abs/2003.12294)) -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| |---|---|---|---|---| diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index d70f99bb5c5b0bdcb7d39209dfc9a77c56918260..a34508f952cb738d4f12c4fec7d0c34e18bab282 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -11,9 +11,10 @@ This tutorial lists the text detection algorithms and text recognition algorithm ### 1. Text Detection Algorithm PaddleOCR open source text detection algorithms list: -- [x] EAST([paper](https://arxiv.org/abs/1704.03155))[2] -- [x] DB([paper](https://arxiv.org/abs/1911.08947))[1] -- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] +- [x] EAST([paper](https://arxiv.org/abs/1704.03155)) +- [x] DB([paper](https://arxiv.org/abs/1911.08947)) +- [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: @@ -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|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)| +|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%|70.47%|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: @@ -41,11 +44,11 @@ For the training guide and use of PaddleOCR text detection algorithms, please re ### 2. Text Recognition Algorithm PaddleOCR open-source text recognition algorithms list: -- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7] -- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] -- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] -- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] -- [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] +- [x] CRNN([paper](https://arxiv.org/abs/1507.05717)) +- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085)) +- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) +- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1)) +- [x] SRN([paper](https://arxiv.org/abs/2003.12294)) Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow: