📄 Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.
修改模型转换说明Readme,增加26国语言rec模型,对应修改tools下的转换脚本。
benjaminwanfc3918ff519d
diff --git a/models/readme.md b/models/readme.mdindex e38500c..e609921 100755--- a/models/readme.md+++ b/models/readme.md@@ -1,93 +1,302 @@## Download models+* from baiduNetDisk:-链接:https://pan.baidu.com/s/1UhDrC7iOMWiYaQiHk3acdw 提取码:yjgv+链接:https://pan.baidu.com/s/1UhDrC7iOMWiYaQiHk3acdw 提取码:yjgv## onnx模型转换说明+### 一、如何获得模型+* 您可以通过使用paddle全家桶,通过其完善的说明,事无巨细的指引,手把手地让您从零开始,训练自己的模型。* 或者,您也可以直接去paddle官网下载预先训练好的模型。* 这个预先训练好的模型叫作train模型,server模型体积较大(准确度高速度慢些),mobile模型体积较小(准确度相对低速度快)。* det和rec有server和mobile两种模型,cls只有mobile模型。* 预训练模型下载地址:-1. [mobile det train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar)-2. [mobile cls train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar)-3. [mobile rec train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar)-4. [server det train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar)-5. [server rec train](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar)++|模型名称|模型简介|配置文件|下载地址|+| --- | --- | --- | --- |+|mobile cls train|文本方向分类|rec_chinese_lite_train_v2.0.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) |+|ch_ppocr_mobile_v2.0_det|超轻量模型,多语种文本检测|ch_det_mv3_db_v2.0.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar)|+|ch_ppocr_server_v2.0_det|通用模型,多语种文本检测|ch_det_res18_db_v2.0.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar)|+|ch_ppocr_mobile_v2.0_rec|超轻量模型,简体中英文数字|rec_chinese_lite_train_v2.0.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) |+|ch_ppocr_server_v2.0_rec|通用模型,简体中英文数字|rec_chinese_common_train_v2.0.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) |+|en_number_mobile_v2.0_rec|英文数字|rec_en_number_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_train.tar) |+|french_mobile_v2.0_rec|法文|rec_french_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/french_mobile_v2.0_rec_train.tar) |+|german_mobile_v2.0_rec|德文|rec_german_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/german_mobile_v2.0_rec_train.tar) |+|korean_mobile_v2.0_rec|韩文|rec_korean_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_train.tar) |+|japan_mobile_v2.0_rec|日文|rec_japan_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/japan_mobile_v2.0_rec_train.tar) |+|it_mobile_v2.0_rec|意大利文|rec_it_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_train.tar) |+|xi_mobile_v2.0_rec|西班牙文|rec_xi_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_train.tar) |+|pu_mobile_v2.0_rec|葡萄牙文|rec_pu_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_train.tar) |+|ru_mobile_v2.0_rec|俄罗斯文|rec_ru_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_train.tar) |+|ar_mobile_v2.0_rec|阿拉伯文|rec_ar_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_train.tar) |+|hi_mobile_v2.0_rec|印地文|rec_hi_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_train.tar) |+|chinese_cht_mobile_v2.0_rec|繁体中文|rec_chinese_cht_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_train.tar) |+|ug_mobile_v2.0_rec|维吾尔文|rec_ug_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_train.tar) |+|fa_mobile_v2.0_rec|波斯文|rec_fa_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_train.tar) |+|ur_mobile_v2.0_rec|乌尔都文|rec_ur_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_train.tar) |+|rs_mobile_v2.0_rec|塞尔维亚文(latin)|rec_rs_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_train.tar) |+|oc_mobile_v2.0_rec|欧西坦文|rec_oc_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_train.tar) |+|mr_mobile_v2.0_rec|马拉地文|rec_mr_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_train.tar) |+|ne_mobile_v2.0_rec|尼泊尔文|rec_ne_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_train.tar) |+|rsc_mobile_v2.0_rec|塞尔维亚文(cyrillic)|rec_rsc_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_train.tar) |+|bg_mobile_v2.0_rec|保加利亚文|rec_bg_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_train.tar) |+|uk_mobile_v2.0_rec|乌克兰文|rec_uk_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_train.tar) |+|be_mobile_v2.0_rec|白俄罗斯文|rec_be_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_train.tar) |+|te_mobile_v2.0_rec|泰卢固文|rec_te_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_train.tar) |+|ka_mobile_v2.0_rec|卡纳达文|rec_ka_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |+|ta_mobile_v2.0_rec|泰米尔文|rec_ta_lite_train.yml|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_train.tar) |+* 先把这些模型通过训练或者下载的方式准备好备用。### 二、准备环境和工具+* 开始之前,需要先准备好Linux,paddle的pip包不支持macOS,而windows部署起来麻烦一些,以下说明以deepin20为范例。* 安装python3: ```sudo apt-get install python3```-* 安装pip3:+* 安装pip3:+```wget https://bootstrap.pypa.io/get-pip.pysudo python3 get-pip.py```-* pip安装paddlepaddle(2.0的模型需要2.0版): ```sudo pip3 install paddlepaddle==2.0.0rc1 -i https://mirror.baidu.com/pypi/simple```-* pip安装转换工具附加依赖: ```sudo pip3 install setuptools tqdm opencv-python imgaug pyclipper -i https://mirror.baidu.com/pypi/simple```++* pip安装paddlepaddle(+ 2.0的模型需要2.0版): ```sudo pip3 install paddlepaddle==2.0.0rc1 -i https://mirror.baidu.com/pypi/simple```+*++pip安装转换工具附加依赖: ```sudo pip3 install setuptools tqdm opencv-python imgaug pyclipper -i https://mirror.baidu.com/pypi/simple```+* 安装lmdb依赖: ```sudo apt-get install python3-yaml libxml2-dev libxslt1-dev python3-dev```* pip安装lmdb: ```sudo pip3 install lmdb -i https://mirror.baidu.com/pypi/simple```* pip安装onnx: ```sudo pip3 install onnx```-* 下载PaddleOCR(用于预训练模型转paddle推理模型):```git clone https://github.com/PaddlePaddle/PaddleOCR.git```+* 下载PaddleOCR(用于预训练模型转paddle推理模型):```git clone https://github.com/PaddlePaddle/PaddleOCR.git```* 安装转换工具库Paddle2ONNX(develop分支,用于paddle推理模型转onnx模型):+```shell-#以下为修复mobile rec转换错误的开发版-#修复代码还未同步进官方仓库之前,先使用此版本-git clone https://github.com/Channingss/paddle2onnx.git-cd paddle2onnx-git checkout fix_adative_pool+#官方仓库已修复fix_adative_pool+git clone https://github.com/PaddlePaddle/Paddle2ONNX.git+cd Paddle2ONNX+git checkout developsudo python3 setup.py install-#以下是官方仓库-#git clone https://github.com/PaddlePaddle/Paddle2ONNX.git-#cd Paddle2ONNX+#以下为修复mobile rec转换错误的开发版+#修复代码还未同步进官方仓库之前,先使用此版本+#git clone https://github.com/Channingss/paddle2onnx.git+#cd paddle2onnx+#git checkout fix_adative_pool#sudo python3 setup.py install```### 三、开始转换+* 接下来的操作都在PaddleOCR文件夹里进行-* PaddleOCR切换到2.0分支(因为预训练模型也是v2.0格式的)+* PaddleOCR切换到release/2.0分支(因为train训练模型也是v2.0格式的)+```shellgit clone https://github.com/PaddlePaddle/PaddleOCR.gitcd PaddleOCR-git checkout release/2.0-rc1-0+git checkout release/2.0```-* 新建文件夹PaddleOCR/models-* 把预训练模型解压到models目录,严格按照如下目录结构和文件名放置++* 生成多语言识别模型配置文件++```shell+#安装模块依赖+sudo pip3 install pyyaml+cd PaddleOCR/configs/rec/multi_language+#生成多语言配置文件+python3 generate_multi_language_configs.py -l it+python3 generate_multi_language_configs.py -l xi+python3 generate_multi_language_configs.py -l pu+python3 generate_multi_language_configs.py -l ru+python3 generate_multi_language_configs.py -l ar+python3 generate_multi_language_configs.py -l hi+python3 generate_multi_language_configs.py -l chinese_cht+python3 generate_multi_language_configs.py -l ug+python3 generate_multi_language_configs.py -l fa+python3 generate_multi_language_configs.py -l ur+python3 generate_multi_language_configs.py -l rs+python3 generate_multi_language_configs.py -l oc+python3 generate_multi_language_configs.py -l mr+python3 generate_multi_language_configs.py -l ne+python3 generate_multi_language_configs.py -l rsc+python3 generate_multi_language_configs.py -l bg+python3 generate_multi_language_configs.py -l uk+python3 generate_multi_language_configs.py -l be+python3 generate_multi_language_configs.py -l te+python3 generate_multi_language_configs.py -l ka+python3 generate_multi_language_configs.py -l ta```-PaddleOCR/models- ├── ch_ppocr_mobile_v2.0_cls_train- │ ├── best_accuracy.pdopt- │ ├── best_accuracy.pdparams- │ ├── best_accuracy.states- │ └── train.log- ├── ch_ppocr_mobile_v2.0_det_train- │ ├── best_accuracy.pdopt- │ ├── best_accuracy.pdparams- │ ├── best_accuracy.states- │ └── train.log- ├── ch_ppocr_mobile_v2.0_rec_train- │ ├── best_accuracy.pdparams- │ └── train.log- ├── ch_ppocr_server_v2.0_det_train- │ ├── best_accuracy.pdopt- │ ├── best_accuracy.pdparams- │ ├── best_accuracy.states- │ └── train.log- ├── ch_ppocr_server_v2.0_rec_train- │ ├── best_accuracy.pdopt- │ ├── best_accuracy.pdparams- │ ├── best_accuracy.states- │ └── train.log++* 新建文件夹PaddleOCR/inference+* 把预训练模型解压到inference目录,严格按照如下目录结构和文件名放置++```+PaddleOCR/inference+├── ar_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── be_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── bg_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ch_ppocr_mobile_v2.0_cls_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ch_ppocr_mobile_v2.0_det_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ch_ppocr_mobile_v2.0_rec_train+│ ├── best_accuracy.pdparams+│ └── train.log+├── ch_ppocr_server_v2.0_det_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ch_ppocr_server_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── chinese_cht_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── en_number_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── fa_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── french_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── german_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── hi_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── it_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── japan_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ka_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── korean_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── mr_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ne_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── oc_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── pu_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── rs_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── rsc_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ru_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ta_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── te_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ug_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── uk_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+├── ur_mobile_v2.0_rec_train+│ ├── best_accuracy.pdopt+│ ├── best_accuracy.pdparams+│ ├── best_accuracy.states+│ └── train.log+└── xi_mobile_v2.0_rec_train+ ├── best_accuracy.pdopt+ ├── best_accuracy.pdparams+ ├── best_accuracy.states+ └── train.log```+* 把本项目tools文件夹里的export_custom.py复制到PaddleOCR/tools里,把program.py覆盖到PaddleOCR/tools里* 把本项目tools文件夹里的gen-models.sh复制到PaddleOCR根目录* 开始转换+```shellcd PaddleOCRchmod a+x gen-models.sh./gen-models.sh```-* 转换成功后,会在PaddleOCR/models文件夹里生成5个onnx模型++* 转换成功后,会在PaddleOCR/inference文件夹里生成31个onnx模型diff --git a/tools/export_custom.py b/tools/export_custom.pyindex c03ad74..6a79867 100755--- a/tools/export_custom.py+++ b/tools/export_custom.py@@ -36,8 +36,6 @@ def main():config = load_config(FLAGS.config)merge_config(FLAGS.opt)logger = get_logger()- logger.info('input height is {}'.format(FLAGS.height))- logger.info('input width is {}'.format(FLAGS.width))# build post processpost_process_class = build_post_process(config['PostProcess'],@@ -53,7 +51,11 @@ def main():model.eval()save_path = '{}/inference'.format(config['Global']['save_inference_dir'])- infer_shape = [3, int(FLAGS.height), int(FLAGS.width)]++ infer_shape = [3, int(FLAGS.height), int(FLAGS.width)]+ if config['Architecture']['model_type'] == "rec":+ infer_shape = [3, 32, -1]+model = to_static(model,input_spec=[diff --git a/tools/gen-models.sh b/tools/gen-models.shindex 900fb3f..f35260e 100755--- a/tools/gen-models.sh+++ b/tools/gen-models.sh@@ -1,32 +1,110 @@#!/bin/bash-# ======ch_ppocr_mobile_v2.0_cls======-python3 tools/export_custom.py --height 48 --width 192 -c configs/cls/cls_mv3.yml -o Global.pretrained_model=./models/ch_ppocr_mobile_v2.0_cls_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./models/ch_ppocr_mobile_v2.0_cls_infer/+function trainToOnnx() {+ python3 tools/export_custom.py --height ${1} \+ --width ${2} \+ -c ${3} \+ -o Global.pretrained_model=./inference/${4}_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/${4}_infer/-paddle2onnx --model_dir models/ch_ppocr_mobile_v2.0_cls_infer/ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file models/ch_ppocr_mobile_v2.0_cls_infer.onnx --opset_version 11 --enable_onnx_checker True+ paddle2onnx --model_dir inference/${4}_infer \+ --model_filename inference.pdmodel \+ --params_filename inference.pdiparams \+ --save_file inference/${4}_infer.onnx \+ --opset_version 11 \+ --enable_onnx_checker True+}-# ======ch_ppocr_mobile_v2.0_det======-python3 tools/export_custom.py --height -1 --width -1 -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./models/ch_ppocr_mobile_v2.0_det_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./models/ch_ppocr_mobile_v2.0_det_infer/+# ch_ppocr_mobile_v2.0_cls+trainToOnnx 48 192 configs/cls/cls_mv3.yml ch_ppocr_mobile_v2.0_cls-paddle2onnx --model_dir models/ch_ppocr_mobile_v2.0_det_infer/ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file models/ch_ppocr_mobile_v2.0_det_infer.onnx --opset_version 11 --enable_onnx_checker True+# ch_ppocr_mobile_v2.0_det+trainToOnnx -1 -1 configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml ch_ppocr_mobile_v2.0_det-# ======ch_ppocr_mobile_v2.0_rec======-python3 tools/export_custom.py --height 32 --width -1 -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./models/ch_ppocr_mobile_v2.0_rec_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./models/ch_ppocr_mobile_v2.0_rec_infer/+# ch_ppocr_server_v2.0_det+trainToOnnx -1 -1 configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml ch_ppocr_server_v2.0_det-paddle2onnx --model_dir models/ch_ppocr_mobile_v2.0_rec_infer/ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file models/ch_ppocr_mobile_v2.0_rec_infer.onnx --opset_version 11 --enable_onnx_checker True+# ch_ppocr_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml ch_ppocr_mobile_v2.0_rec+# ch_ppocr_server_v2.0_rec+trainToOnnx 32 -1 configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml ch_ppocr_server_v2.0_rec-# ======ch_ppocr_server_v2.0_det======-python3 tools/export_custom.py --height -1 --width -1 -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o Global.pretrained_model=./models/ch_ppocr_server_v2.0_det_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./models/ch_ppocr_server_v2.0_det_infer/+# en_number_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_en_number_lite_train.yml en_number_mobile_v2.0_rec-paddle2onnx --model_dir models/ch_ppocr_server_v2.0_det_infer/ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file models/ch_ppocr_server_v2.0_det_infer.onnx --opset_version 11 --enable_onnx_checker True+# french_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_french_lite_train.yml french_mobile_v2.0_rec+# german_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_german_lite_train.yml german_mobile_v2.0_rec-# ======ch_ppocr_server_v2.0_rec======-python3 tools/export_custom.py --height 32 --width -1 -c configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml -o Global.pretrained_model=./models/ch_ppocr_server_v2.0_rec_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./models/ch_ppocr_server_v2.0_rec_infer/+# korean_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_korean_lite_train.yml korean_mobile_v2.0_rec-paddle2onnx --model_dir models/ch_ppocr_server_v2.0_rec_infer/ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file models/ch_ppocr_server_v2.0_rec_infer.onnx --opset_version 11 --enable_onnx_checker True+# japan_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_japan_lite_train.yml japan_mobile_v2.0_rec+# it_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_it_lite_train.yml it_mobile_v2.0_rec+# xi_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_xi_lite_train.yml xi_mobile_v2.0_rec++# pu_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_pu_lite_train.yml pu_mobile_v2.0_rec++# ru_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ru_lite_train.yml ru_mobile_v2.0_rec++# ar_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ar_lite_train.yml ar_mobile_v2.0_rec++# hi_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_hi_lite_train.yml hi_mobile_v2.0_rec++# chinese_cht_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_chinese_cht_lite_train.yml chinese_cht_mobile_v2.0_rec++# ug_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ug_lite_train.yml ug_mobile_v2.0_rec++# fa_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_fa_lite_train.yml fa_mobile_v2.0_rec++# ur_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ur_lite_train.yml ur_mobile_v2.0_rec++# rs_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_rs_lite_train.yml rs_mobile_v2.0_rec++# oc_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_oc_lite_train.yml oc_mobile_v2.0_rec++# mr_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_mr_lite_train.yml mr_mobile_v2.0_rec++# ne_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ne_lite_train.yml ne_mobile_v2.0_rec++# rsc_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_rsc_lite_train.yml rsc_mobile_v2.0_rec++# bg_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_bg_lite_train.yml bg_mobile_v2.0_rec++# uk_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_uk_lite_train.yml uk_mobile_v2.0_rec++# be_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_be_lite_train.yml be_mobile_v2.0_rec++# te_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_te_lite_train.yml te_mobile_v2.0_rec++# ka_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ka_lite_train.yml ka_mobile_v2.0_rec++# ta_mobile_v2.0_rec+trainToOnnx 32 -1 configs/rec/multi_language/rec_ta_lite_train.yml ta_mobile_v2.0_recdiff --git a/tools/program.py b/tools/program.pyindex 802365b..b8c94db 100755--- a/tools/program.py+++ b/tools/program.py@@ -133,7 +133,7 @@ def check_gpu(use_gpu):"model on CPU"try:- if use_gpu and not paddle.fluid.is_compiled_with_cuda():+ if use_gpu and not paddle.is_compiled_with_cuda():print(err)sys.exit(1)except Exception as e:@@ -181,11 +181,11 @@ def train(config,if 'start_epoch' in best_model_dict:start_epoch = best_model_dict['start_epoch']else:- start_epoch = 0+ start_epoch = 1- for epoch in range(start_epoch, epoch_num):- if epoch > 0:- train_dataloader = build_dataloader(config, 'Train', device, logger)+ for epoch in range(start_epoch, epoch_num + 1):+ train_dataloader = build_dataloader(+ config, 'Train', device, logger, seed=epoch)train_batch_cost = 0.0train_reader_cost = 0.0batch_sum = 0@@ -214,7 +214,7 @@ def train(config,stats['lr'] = lrtrain_stats.update(stats)- if cal_metric_during_train: # onlt rec and cls need+ if cal_metric_during_train: # only rec and cls needbatch = [item.numpy() for item in batch]post_result = post_process_class(preds, batch[1])eval_class(post_result, batch)@@ -240,21 +240,21 @@ def train(config,# evalif global_step > start_eval_step and \(global_step - start_eval_step) % eval_batch_step == 0 and dist.get_rank() == 0:- cur_metirc = eval(model, valid_dataloader, post_process_class,+ cur_metric = eval(model, valid_dataloader, post_process_class,eval_class)- cur_metirc_str = 'cur metirc, {}'.format(', '.join(- ['{}: {}'.format(k, v) for k, v in cur_metirc.items()]))- logger.info(cur_metirc_str)+ cur_metric_str = 'cur metric, {}'.format(', '.join(+ ['{}: {}'.format(k, v) for k, v in cur_metric.items()]))+ logger.info(cur_metric_str)# logger metricif vdl_writer is not None:- for k, v in cur_metirc.items():+ for k, v in cur_metric.items():if isinstance(v, (float, int)):vdl_writer.add_scalar('EVAL/{}'.format(k),- cur_metirc[k], global_step)- if cur_metirc[main_indicator] >= best_model_dict[+ cur_metric[k], global_step)+ if cur_metric[main_indicator] >= best_model_dict[main_indicator]:- best_model_dict.update(cur_metirc)+ best_model_dict.update(cur_metric)best_model_dict['best_epoch'] = epochsave_model(model,@@ -265,7 +265,7 @@ def train(config,prefix='best_accuracy',best_model_dict=best_model_dict,epoch=epoch)- best_str = 'best metirc, {}'.format(', '.join([+ best_str = 'best metric, {}'.format(', '.join(['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))logger.info(best_str)@@ -296,7 +296,7 @@ def train(config,prefix='iter_epoch_{}'.format(epoch),best_model_dict=best_model_dict,epoch=epoch)- best_str = 'best metirc, {}'.format(', '.join(+ best_str = 'best metric, {}'.format(', '.join(['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))logger.info(best_str)if dist.get_rank() == 0 and vdl_writer is not None:@@ -325,13 +325,13 @@ def eval(model, valid_dataloader, post_process_class, eval_class):eval_class(post_result, batch)pbar.update(1)total_frame += len(images)- # Get final metirc,eg. acc or hmean- metirc = eval_class.get_metric()+ # Get final metric,eg. acc or hmean+ metric = eval_class.get_metric()pbar.close()model.train()- metirc['fps'] = total_frame / total_time- return metirc+ metric['fps'] = total_frame / total_time+ return metricdef preprocess(is_train=False):