在全新ubuntu上用gpu训练paddleocr模型遇到的坑与解决办法

一. 我的ubuntu版本

二.首先拉取paddleocr源代码

下载地址：https://gitee.com/paddlepaddle/PaddleOCR

三.下载模型

• 我要训练一个中文模型，看到该预训练模型泛化性能最优，于是下载这个模型
  https://gitee.com/link?target=https%3A%2F%2Fpaddleocr.bj.bcebos.com%2FPP-OCRv3%2Fchinese%2Fch_PP-OCRv3_rec_train.tar

• 其他模型地址：https://gitee.com/paddlepaddle/PaddleOCR/blob/release/2.6/doc/doc_ch/models_list.md

四.训练前的准备

1.在源代码文件夹里创造一个自己放东西的文件

• config文件夹用来装yml配置文件
  pretrained_model用来装上一步下载的预训练模型
  split_rec_label用来放数据集
  output用来放训练出的模型

• 创建文件夹非强制，只是这样更方便管理自己文件，yml源文件地址就在
  PaddleOCR-release-2.6/configs/rec/PP-OCRv3这个路径下

2.准备数据

2.1数据标注

参考博客：https://blog.csdn.net/qq_49627063/article/details/119134847

2.2数据划分

在训练之前，所有图片都在一个文件夹中，所有label信息都在同一个txt文件中，因此需要编写脚本，将其按照8:1:1的比例进行分割。

import os
import re
import shutil
import random
import argparsedef split_label(all_label, train_label, val_label, test_label):f = open(all_label, 'r')f_train = open(train_label, 'w')f_val = open(val_label, 'w')f_test = open(test_label, 'w')raw_list = f.readlines()num_train = int(len(raw_list) * 0.8)num_val = int(len(raw_list) * 0.1)num_test = int(len(raw_list) * 0.1)random.shuffle(raw_list)for i in range(num_train):f_train.writelines(raw_list[i])for i in range(num_train, num_train + num_val):f_val.writelines(raw_list[i])for i in range(num_train + num_val, num_train + num_val + num_test):f_test.writelines(raw_list[i])f.close()f_train.close()f_val.close()f_test.close()def split_img(all_imgs, train_label, train_imgs, val_label, val_imgs, test_label, test_imgs):f_train = open(train_label, 'r')f_val = open(val_label, 'r')f_test = open(test_label, 'r')train_list = f_train.readlines()val_list = f_val.readlines()test_list = f_test.readlines()for i in range(len(train_list)):img_path = os.path.join(all_imgs, re.split("[/\t]", train_list[i])[1])shutil.move(img_path, train_imgs)for i in range(len(val_list)):img_path = os.path.join(all_imgs, re.split("[/\t]", val_list[i])[1])shutil.move(img_path, val_imgs)for i in range(len(test_list)):img_path = os.path.join(all_imgs, re.split("[/\t]", test_list[i])[1])shutil.move(img_path, test_imgs)def get_args():parser = argparse.ArgumentParser()parser.add_argument("--all_label", default="../paddleocr/PaddleOCR/train_data/cls/cls_gt_train.txt")parser.add_argument("--all_imgs_dir", default="../paddleocr/PaddleOCR/train_data/cls/images/")parser.add_argument("--train_label", default="../paddleocr/PaddleOCR/train_data/cls/train.txt")parser.add_argument("--train_imgs_dir", default="../paddleocr/PaddleOCR/train_data/cls/train/")parser.add_argument("--val_label", default="../paddleocr/PaddleOCR/train_data/cls/val.txt")parser.add_argument("--val_imgs_dir", default="../paddleocr/PaddleOCR/train_data/cls/val/")parser.add_argument("--test_label", default="../paddleocr/PaddleOCR/train_data/cls/test.txt")parser.add_argument("--test_imgs_dir", default="../paddleocr/PaddleOCR/train_data/cls/test/")return parser.parse_args()def main(args):if not os.path.isdir(args.train_imgs_dir):os.makedirs(args.train_imgs_dir)if not os.path.isdir(args.val_imgs_dir):os.makedirs(args.val_imgs_dir)if not os.path.isdir(args.test_imgs_dir):os.makedirs(args.test_imgs_dir)split_label(args.all_label, args.train_label, args.val_label, args.test_label)split_img(args.all_imgs_dir, args.train_label, args.train_imgs_dir, args.val_label, args.val_imgs_dir, args.test_label, args.test_imgs_dir)if __name__ == "__main__":main(get_args())

3.改写yml配置文件

• 源地址：https://gitee.com/paddlepaddle/PaddleOCR/blob/release/2.6/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml

Global:debug: falseuse_gpu: trueepoch_num: 800log_smooth_window: 20print_batch_step: 10save_model_dir: wjp/output/rec_ppocr_v3_distillationsave_epoch_step: 3eval_batch_step: [0, 2000]cal_metric_during_train: truepretrained_model:checkpoints:save_inference_dir:use_visualdl: falseinfer_img: doc/imgs_words/ch/word_1.jpgcharacter_dict_path: ppocr/utils/ppocr_keys_v1.txtmax_text_length: &max_text_length 25infer_mode: falseuse_space_char: truedistributed: truesave_res_path: wjp/output/rec/predicts_ppocrv3_distillation.txtOptimizer:name: Adambeta1: 0.9beta2: 0.999lr:name: Piecewisedecay_epochs : [700]values : [0.0005, 0.00005]warmup_epoch: 5regularizer:name: L2factor: 3.0e-05Architecture:model_type: &model_type "rec"name: DistillationModelalgorithm: DistillationModels:Teacher:pretrained:freeze_params: falsereturn_all_feats: truemodel_type: *model_typealgorithm: SVTRTransform:Backbone:name: MobileNetV1Enhancescale: 0.5last_conv_stride: [1, 2]last_pool_type: avgHead:name: MultiHeadhead_list:- CTCHead:Neck:name: svtrdims: 64depth: 2hidden_dims: 120use_guide: TrueHead:fc_decay: 0.00001- SARHead:enc_dim: 512max_text_length: *max_text_lengthStudent:pretrained:freeze_params: falsereturn_all_feats: truemodel_type: *model_typealgorithm: SVTRTransform:Backbone:name: MobileNetV1Enhancescale: 0.5last_conv_stride: [1, 2]last_pool_type: avgHead:name: MultiHeadhead_list:- CTCHead:Neck:name: svtrdims: 64depth: 2hidden_dims: 120use_guide: TrueHead:fc_decay: 0.00001- SARHead:enc_dim: 512max_text_length: *max_text_length
Loss:name: CombinedLossloss_config_list:- DistillationDMLLoss:weight: 1.0act: "softmax"use_log: truemodel_name_pairs:- ["Student", "Teacher"]key: head_outmulti_head: Truedis_head: ctcname: dml_ctc- DistillationDMLLoss:weight: 0.5act: "softmax"use_log: truemodel_name_pairs:- ["Student", "Teacher"]key: head_outmulti_head: Truedis_head: sarname: dml_sar- DistillationDistanceLoss:weight: 1.0mode: "l2"model_name_pairs:- ["Student", "Teacher"]key: backbone_out- DistillationCTCLoss:weight: 1.0model_name_list: ["Student", "Teacher"]key: head_outmulti_head: True- DistillationSARLoss:weight: 1.0model_name_list: ["Student", "Teacher"]key: head_outmulti_head: TruePostProcess:name: DistillationCTCLabelDecodemodel_name: ["Student", "Teacher"]key: head_outmulti_head: TrueMetric:name: DistillationMetricbase_metric_name: RecMetricmain_indicator: acckey: "Student"ignore_space: FalseTrain:dataset:name: SimpleDataSetdata_dir: wjp/split_rec_label/trainext_op_transform_idx: 1label_file_list:- wjp/split_rec_label/train.txttransforms:- DecodeImage:img_mode: BGRchannel_first: false- RecConAug:prob: 0.5ext_data_num: 2image_shape: [48, 320, 3]max_text_length: *max_text_length- RecAug:- MultiLabelEncode:- RecResizeImg:image_shape: [3, 48, 320]- KeepKeys:keep_keys:- image- label_ctc- label_sar- length- valid_ratioloader:shuffle: truebatch_size_per_card: 32drop_last: truenum_workers: 4
Eval:dataset:name: SimpleDataSetdata_dir: wjp/split_rec_label/vallabel_file_list:- wjp/split_rec_label/val.txttransforms:- DecodeImage:img_mode: BGRchannel_first: false- MultiLabelEncode:- RecResizeImg:image_shape: [3, 48, 320]- KeepKeys:keep_keys:- image- label_ctc- label_sar- length- valid_ratioloader:shuffle: falsedrop_last: falsebatch_size_per_card: 128num_workers: 4

4.安装anaconda

参考博客：https://blog.csdn.net/wyf2017/article/details/118676765

• 创建python虚拟环境

conda create -n ppocr

• 切换虚拟环境

source activate ppocr

五.开始训练

python tools/train.py -c wjp/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model=wjp/ch_PP-OCRv3_rec_train/best_accuracy
//-c参数放配置文件地址，-o参数放预训练模型地址

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple

六.报错

（1） libGL.so.1

ImportError: libGL.so.1: cannot open shared object file: No such file or directory

• 解决办法：

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv-python-headless

（2）Polygon

ModuleNotFoundError: No module named 'Polygon'

• 解决办法：

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple Polygon3

(3) lanms

ModuleNotFoundError: No module named 'lanms'

源码下载地址：https://github.com/AndranikSargsyan/lanms-nova/tree/master

参考我这个教程编译：http://t.csdnimg.cn/BqOW6

• 将__init __.py文件替换

import numpy as npdef merge_quadrangle_n9(polys, thres=0.3, precision=10000):if len(polys) == 0:return np.array([], dtype='float32')p = polys.copy()p[:, :8] *= precisionret = np.array(merge_quadrangle_n9(p, thres), dtype='float32')ret[:, :8] /= precisionreturn ret

• 找到linux种anaconda的包放在什么地方

pip show numpy

就知道该环境下的包安装地址

• 将编译好库的整个lanms文件夹移动到该地址去即可调用

（4）报错UnicodeDecodeError: ‘utf-8’ codec can’t decode byte 0xbc in position 2: invalid start byt

f = open('txt01.txt',encoding='utf-8')

将 encoding=’utf-8’ 改为GB2312、gbk、ISO-8859-1，随便尝试一个均可以

（5）Out of memory error on GPU 0. Cannot allocate xxxxMB memory on GPU 0, xxxxGB memory has been allocated and available memory is only 0.000000B.

将训练的配置yml文件中的batch_size_per_card参数不断改小（除以2），直到不再报这个错即可。