From c269e1e42f79d33aa87d5e0ba49b488c56f97ae4 Mon Sep 17 00:00:00 2001 From: Topdu <784990967@qq.com> Date: Mon, 30 May 2022 08:36:11 +0000 Subject: [PATCH] modify vitstr name --- configs/rec/rec_mtb_nrtr.yml | 2 +- ...{rec_vitstr.yml => rec_vitstr_none_ce.yml} | 4 +-- doc/doc_ch/algorithm_overview.md | 2 +- doc/doc_ch/algorithm_rec_vitstr.md | 26 +++++++++---------- doc/doc_en/algorithm_overview_en.md | 2 +- doc/doc_en/algorithm_rec_vitstr_en.md | 14 +++++----- ppocr/losses/__init__.py | 7 +++-- .../{rec_ce_smooth_loss.py => rec_ce_loss.py} | 4 +-- .../configs/rec_mtb_nrtr/rec_mtb_nrtr.yml | 2 +- .../rec_mtb_nrtr/train_infer_python.txt | 4 +-- .../rec_vitstr_none_ce.yml} | 4 +-- .../train_infer_python.txt | 10 +++---- 12 files changed, 40 insertions(+), 41 deletions(-) rename configs/rec/{rec_vitstr.yml => rec_vitstr_none_ce.yml} (97%) rename ppocr/losses/{rec_ce_smooth_loss.py => rec_ce_loss.py} (93%) rename test_tipc/configs/{rec_vitstr/rec_vitstr.yml => rec_vitstr_none_ce/rec_vitstr_none_ce.yml} (97%) rename test_tipc/configs/{rec_vitstr => rec_vitstr_none_ce}/train_infer_python.txt (82%) diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml index dfe2cc98..40567488 100644 --- a/configs/rec/rec_mtb_nrtr.yml +++ b/configs/rec/rec_mtb_nrtr.yml @@ -49,7 +49,7 @@ Architecture: Loss: - name: CESmoothingLoss + name: CELoss smoothing: True PostProcess: diff --git a/configs/rec/rec_vitstr.yml b/configs/rec/rec_vitstr_none_ce.yml similarity index 97% rename from configs/rec/rec_vitstr.yml rename to configs/rec/rec_vitstr_none_ce.yml index 005db018..065cc785 100644 --- a/configs/rec/rec_vitstr.yml +++ b/configs/rec/rec_vitstr_none_ce.yml @@ -3,7 +3,7 @@ Global: epoch_num: 20 log_smooth_window: 20 print_batch_step: 10 - save_model_dir: ./output/rec/vitstr/ + save_model_dir: ./output/rec/vitstr_none_ce/ save_epoch_step: 1 # evaluation is run every 2000 iterations after the 0th iteration# eval_batch_step: [0, 50] @@ -44,7 +44,7 @@ Architecture: name: CTCHead Loss: - name: CESmoothingLoss + name: CELoss smoothing: False with_all: True diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index 934ac085..4196ec05 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -85,7 +85,7 @@ |SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) | |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) | |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) | -|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) | +|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) | diff --git a/doc/doc_ch/algorithm_rec_vitstr.md b/doc/doc_ch/algorithm_rec_vitstr.md index bd83b8d9..ab12be72 100644 --- a/doc/doc_ch/algorithm_rec_vitstr.md +++ b/doc/doc_ch/algorithm_rec_vitstr.md @@ -27,7 +27,7 @@ |模型|骨干网络|配置文件|Acc|下载链接| | --- | --- | --- | --- | --- | -|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)| +|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar)| ## 2. 环境配置 @@ -40,7 +40,7 @@ ### 3.1 模型训练 -请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_ViTSTR.yml)。 +请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_vitstr_none_ce.yml)。 #### 启动训练 @@ -48,10 +48,10 @@ 具体地，在完成数据准备后，便可以启动训练，训练命令如下： ```shell #单卡训练（训练周期长，不建议） -python3 tools/train.py -c configs/rec/rec_vitstr.yml +python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml #多卡训练，通过--gpus参数指定卡号 -python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_vitstr.yml +python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_vitstr_none_ce.yml ``` @@ -61,7 +61,7 @@ python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs ```shell # 注意将pretrained_model的路径设置为本地路径。 -python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy +python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy ``` @@ -70,7 +70,7 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec 使用如下命令进行单张图片预测： ```shell # 注意将pretrained_model的路径设置为本地路径。 -python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy +python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy # 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。 ``` @@ -80,15 +80,15 @@ python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./ ### 4.1 Python推理 -首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) )，可以使用如下命令进行转换： +首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) )，可以使用如下命令进行转换： ```shell # 注意将pretrained_model的路径设置为本地路径。 -python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/ +python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/ ``` **注意：** - 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。 -- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应NRTR的`infer_shape`。 +- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ViTSTR的`infer_shape`。 转换成功后，在目录下有三个文件： ``` @@ -110,20 +110,20 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' 执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下： 结果如下： ```shell -Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901) +Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617) ``` **注意**： - 训练上述模型采用的图像分辨率是[1，224，224]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。 - 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。 -- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中NRTR的预处理为您的预处理方法。 +- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ViTSTR的预处理为您的预处理方法。 ### 4.2 C++推理部署 -由于C++预处理后处理还未支持NRTR，所以暂未支持 +由于C++预处理后处理还未支持ViTSTR，所以暂未支持 ### 4.3 Serving服务化部署 @@ -139,7 +139,7 @@ Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901) ## 5. FAQ 1. 在`ViTSTR`论文中，使用在ImageNet1k上的预训练权重进行初始化训练，我们在训练未采用预训练权重，最终精度没有变化甚至有所提高。 -2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果有需要使用small、base版本，可直接使用源开源repo中的预训练权重转为Paddle权重即可使用。 +2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果需要使用small、base版本，可将[ViTSTR源repo](https://github.com/roatienza/deep-text-recognition-benchmark) 中的预训练权重转为Paddle权重使用。 ## 引用 diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index 213d9580..f2284542 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -84,7 +84,7 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r |SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) | |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) | |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) | -|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) | +|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) | diff --git a/doc/doc_en/algorithm_rec_vitstr_en.md b/doc/doc_en/algorithm_rec_vitstr_en.md index 666798c3..b6f26d39 100644 --- a/doc/doc_en/algorithm_rec_vitstr_en.md +++ b/doc/doc_en/algorithm_rec_vitstr_en.md @@ -25,7 +25,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval |Model|Backbone|config|Acc|Download link| | --- | --- | --- | --- | --- | -|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)| +|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)| ## 2. Environment @@ -43,24 +43,24 @@ Specifically, after the data preparation is completed, the training can be start ``` #Single GPU training (long training period, not recommended) -python3 tools/train.py -c configs/rec/rec_vitstr.yml +python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml #Multi GPU training, specify the gpu number through the --gpus parameter -python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_vitstr.yml +python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_vitstr_none_ce.yml ``` Evaluation: ``` # GPU evaluation -python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model={path/to/weights}/best_accuracy +python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model={path/to/weights}/best_accuracy ``` Prediction: ``` # The configuration file used for prediction must match the training -python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy +python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy ``` @@ -71,7 +71,7 @@ python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./ First, the model saved during the ViTSTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)) ), you can use the following command to convert: ``` -python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr +python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr ``` **Note:** @@ -98,7 +98,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows: The result is as follows: ```shell -Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901) +Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617) ``` diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py index 6c4545eb..7bea87f6 100755 --- a/ppocr/losses/__init__.py +++ b/ppocr/losses/__init__.py @@ -30,7 +30,7 @@ from .det_fce_loss import FCELoss from .rec_ctc_loss import CTCLoss from .rec_att_loss import AttentionLoss from .rec_srn_loss import SRNLoss -from .rec_ce_smooth_loss import CESmoothingLoss +from .rec_ce_loss import CELoss from .rec_sar_loss import SARLoss from .rec_aster_loss import AsterLoss from .rec_pren_loss import PRENLoss @@ -60,9 +60,8 @@ def build_loss(config): support_dict = [ 'DBLoss', 'PSELoss', 'EASTLoss', 'SASTLoss', 'FCELoss', 'CTCLoss', 'ClsLoss', 'AttentionLoss', 'SRNLoss', 'PGLoss', 'CombinedLoss', - 'CESmoothingLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', - 'SDMGRLoss', 'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', - 'MultiLoss' + 'CELoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss', + 'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss' ] config = copy.deepcopy(config) module_name = config.pop('name') diff --git a/ppocr/losses/rec_ce_smooth_loss.py b/ppocr/losses/rec_ce_loss.py similarity index 93% rename from ppocr/losses/rec_ce_smooth_loss.py rename to ppocr/losses/rec_ce_loss.py index 22243ed4..b837ac27 100644 --- a/ppocr/losses/rec_ce_smooth_loss.py +++ b/ppocr/losses/rec_ce_loss.py @@ -3,9 +3,9 @@ from paddle import nn import paddle.nn.functional as F -class CESmoothingLoss(nn.Layer): +class CELoss(nn.Layer): def __init__(self, smoothing=True, with_all=False, **kwargs): - super(CESmoothingLoss, self).__init__() + super(CELoss, self).__init__() self.loss_func = nn.CrossEntropyLoss(reduction='mean', ignore_index=0) self.smoothing = smoothing self.with_all = with_all diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml index 3936ab58..ba6728d8 100644 --- a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml +++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml @@ -49,7 +49,7 @@ Architecture: Loss: - name: CESmoothingLoss + name: CELoss smoothing: True PostProcess: diff --git a/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt index de6de5a0..20d984e3 100644 --- a/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt +++ b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt @@ -40,10 +40,10 @@ infer_quant:False inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,32,100" --rec_algorithm="NRTR" --use_gpu:True|False --enable_mkldnn:True|False ---cpu_threads:1|6 +--cpu_threads:6 --rec_batch_num:1|6 --use_tensorrt:True|False ---precision:fp32|int8 +--precision:fp32 --rec_model_dir: --image_dir:./inference/rec_inference --save_log_path:./test/output/ diff --git a/test_tipc/configs/rec_vitstr/rec_vitstr.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml similarity index 97% rename from test_tipc/configs/rec_vitstr/rec_vitstr.yml rename to test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml index 427bce4b..ccd93c9e 100644 --- a/test_tipc/configs/rec_vitstr/rec_vitstr.yml +++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml @@ -3,7 +3,7 @@ Global: epoch_num: 20 log_smooth_window: 20 print_batch_step: 10 - save_model_dir: ./output/rec/vitstr/ + save_model_dir: ./output/rec/vitstr_none_ce/ save_epoch_step: 1 # evaluation is run every 2000 iterations after the 0th iteration# eval_batch_step: [0, 2000] @@ -43,7 +43,7 @@ Architecture: name: CTCHead Loss: - name: CESmoothingLoss + name: CELoss smoothing: False with_all: True diff --git a/test_tipc/configs/rec_vitstr/train_infer_python.txt b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt similarity index 82% rename from test_tipc/configs/rec_vitstr/train_infer_python.txt rename to test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt index 8b58e220..04c5742e 100644 --- a/test_tipc/configs/rec_vitstr/train_infer_python.txt +++ b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt @@ -13,7 +13,7 @@ train_infer_img_dir:./inference/rec_inference null:null ## trainer:norm_train -norm_train:tools/train.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o +norm_train:tools/train.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o pact_train:null fpgm_train:null distill_train:null @@ -21,21 +21,21 @@ null:null null:null ## ===========================eval_params=========================== -eval:tools/eval.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o +eval:tools/eval.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o null:null ## ===========================infer_params=========================== Global.save_inference_dir:./output/ Global.checkpoints: -norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o +norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o quant_export:null fpgm_export:null distill_export:null export1:null export2:null ## -train_model:./inference/rec_vitstr_train/best_accuracy -infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o +train_model:./inference/rec_vitstr_none_ce_train/best_accuracy +infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o infer_quant:False inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR" --use_gpu:True|False -- GitLab