在全新ubuntu上用gpu训练paddleocr模型遇到的坑与解决办法

雪回2023-10-28 13:34

目录

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

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。

python 复制代码 
import os
import re
import shutil
import random
import argparse

def 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

bash 复制代码 
Global:
  debug: false
  use_gpu: true
  epoch_num: 800
  log_smooth_window: 20
  print_batch_step: 10
  save_model_dir: wjp/output/rec_ppocr_v3_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
  max_text_length: &max_text_length 25
  infer_mode: false
  use_space_char: true
  distributed: true
  save_res_path: wjp/output/rec/predicts_ppocrv3_distillation.txt


Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    name: Piecewise
    decay_epochs : [700]
    values : [0.0005, 0.00005]
    warmup_epoch: 5
  regularizer:
    name: L2
    factor: 3.0e-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: SVTR
      Transform:
      Backbone:
        name: MobileNetV1Enhance
        scale: 0.5
        last_conv_stride: [1, 2]
        last_pool_type: avg
      Head:
        name: MultiHead
        head_list:
          - CTCHead:
              Neck:
                name: svtr
                dims: 64
                depth: 2
                hidden_dims: 120
                use_guide: True
              Head:
                fc_decay: 0.00001
          - SARHead:
              enc_dim: 512
              max_text_length: *max_text_length
    Student:
      pretrained:
      freeze_params: false
      return_all_feats: true
      model_type: *model_type
      algorithm: SVTR
      Transform:
      Backbone:
        name: MobileNetV1Enhance
        scale: 0.5
        last_conv_stride: [1, 2]
        last_pool_type: avg
      Head:
        name: MultiHead
        head_list:
          - CTCHead:
              Neck:
                name: svtr
                dims: 64
                depth: 2
                hidden_dims: 120
                use_guide: True
              Head:
                fc_decay: 0.00001
          - SARHead:
              enc_dim: 512
              max_text_length: *max_text_length
Loss:
  name: CombinedLoss
  loss_config_list:
  - DistillationDMLLoss:
      weight: 1.0
      act: "softmax"
      use_log: true
      model_name_pairs:
      - ["Student", "Teacher"]
      key: head_out
      multi_head: True
      dis_head: ctc
      name: dml_ctc
  - DistillationDMLLoss:
      weight: 0.5
      act: "softmax"
      use_log: true
      model_name_pairs:
      - ["Student", "Teacher"]
      key: head_out
      multi_head: True
      dis_head: sar
      name: dml_sar
  - DistillationDistanceLoss:
      weight: 1.0
      mode: "l2"
      model_name_pairs:
      - ["Student", "Teacher"]
      key: backbone_out
  - DistillationCTCLoss:
      weight: 1.0
      model_name_list: ["Student", "Teacher"]
      key: head_out
      multi_head: True
  - DistillationSARLoss:
      weight: 1.0
      model_name_list: ["Student", "Teacher"]
      key: head_out
      multi_head: True

PostProcess:
  name: DistillationCTCLabelDecode
  model_name: ["Student", "Teacher"]
  key: head_out
  multi_head: True

Metric:
  name: DistillationMetric
  base_metric_name: RecMetric
  main_indicator: acc
  key: "Student"
  ignore_space: False

Train:
  dataset:
    name: SimpleDataSet
    data_dir: wjp/split_rec_label/train
    ext_op_transform_idx: 1
    label_file_list:
    - wjp/split_rec_label/train.txt
    transforms:
    - DecodeImage:
        img_mode: BGR
        channel_first: false
    - RecConAug:
        prob: 0.5
        ext_data_num: 2
        image_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_ratio
  loader:
    shuffle: true
    batch_size_per_card: 32
    drop_last: true
    num_workers: 4
Eval:
  dataset:
    name: SimpleDataSet
    data_dir: wjp/split_rec_label/val
    label_file_list:
    - wjp/split_rec_label/val.txt
    transforms:
    - DecodeImage:
        img_mode: BGR
        channel_first: false
    - MultiLabelEncode:
    - RecResizeImg:
        image_shape: [3, 48, 320]
    - KeepKeys:
        keep_keys:
        - image
        - label_ctc
        - label_sar
        - length
        - valid_ratio
  loader:
    shuffle: false
    drop_last: false
    batch_size_per_card: 128
    num_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
python 复制代码 
conda create -n ppocr
  • 切换虚拟环境
python 复制代码 
source activate ppocr

五.开始训练

python 复制代码 
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

bash 复制代码 
ImportError: libGL.so.1: cannot open shared object file: No such file or directory
  • 解决办法：
bash 复制代码 
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv-python-headless

（2）Polygon

bash 复制代码 
ModuleNotFoundError: No module named 'Polygon'
  • 解决办法：
bash 复制代码 
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple Polygon3

(3) lanms

bash 复制代码 
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
python 复制代码 
import numpy as np


def merge_quadrangle_n9(polys, thres=0.3, precision=10000):
    if len(polys) == 0:
        return np.array([], dtype='float32')
    p = polys.copy()
    p[:, :8] *= precision
    ret = np.array(merge_quadrangle_n9(p, thres), dtype='float32')
    ret[:, :8] /= precision
    return ret
  • 找到linux种anaconda的包放在什么地方
bash 复制代码 
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

bash 复制代码 
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。

相关推荐
神即道 道法自然 如来4 分钟前
Jenkins怎么设置每日自动执行构建任务？
运维·jenkins
小黑爱编程11 分钟前
【LInux】HTTPS是如何实现安全传输的
linux·安全·https
BeyondESH16 分钟前
Linux线程同步—竞态条件和互斥锁(C语言)
linux·服务器·c++
hanniuniu1319 分钟前
详细解读，F5服务器负载均衡的技术优势
运维·服务器·负载均衡
鱼饼6号32 分钟前
Prometheus 上手指南
linux·运维·centos·prometheus
Asher Gu38 分钟前
Linux系统编程入门 | 模拟实现 ls -l 命令
linux
c无序1 小时前
【Linux进程控制】进程程序替换
linux
读书,代码与猫2 小时前
【裸机装机系列】10.kali(ubuntu)-安装nvidia独立显卡步骤
ubuntu·kali·nvidia·linux安装nvidia显卡·ubuntu安装nvidia·裸机安装linux
m0_609000422 小时前
向日葵好用吗？4款稳定的远程控制软件推荐。
运维·服务器·网络·人工智能·远程工作
小安运维日记3 小时前
Linux云计算 |【第四阶段】NOSQL-DAY1
linux·运维·redis·sql·云计算·nosql
热门推荐
01RAG 实践- Ollama+RagFlow 部署本地知识库02组基轨迹建模 GBTM的介绍与实现（Stata 或 R）032024年高教社杯数学建模国赛C题超详细解题思路分析04Coze扣子平台完整体验和实践（附国内和国际版对比）05苍穹外卖面试总结06【2024数模国赛赛题思路公开】国赛B题思路丨附可运行代码丨无偿自提07yolov8实战第五天——yolov8+ffmpg实时视频流检测并进行实时推流——（推流，保姆教学）08CCF-CSP认证考试 202406-3 文本分词 100分题解09【2024高教社杯全国大学生数学建模竞赛】B题 生产过程中的决策问题——解题思路 代码 论文10浏览器插件——ModHeader 的分享和学习