Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

遇见ocrv4训练问题 #11149

Open
yang-chenyu104 opened this issue Oct 27, 2023 · 11 comments
Open

遇见ocrv4训练问题 #11149

yang-chenyu104 opened this issue Oct 27, 2023 · 11 comments
Assignees
Labels
help wanted this issue needs help training this is a training related issue triaged this issue has been looked, and triaged.

Comments

@yang-chenyu104
Copy link

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

  • 系统环境/System Environment:
  • 版本号/Version:Paddle: PaddleOCR: 问题相关组件/Related components:
  • 运行指令/Command Code:
  • 完整报错/Complete Error Message:

我们提供了AceIssueSolver来帮助你解答问题,你是否想要它来解答(请填写yes/no)?/We provide AceIssueSolver to solve issues, do you want it? (Please write yes/no):

请尽量不要包含图片在问题中/Please try to not include the image in the issue.
训练ocrv4看见配置文件中PostProcess方式为蒸馏的CTCLabel,但前处理方式为NRTR导致后续out_channels_list没有找到NRTRLabelDecode,我将编码改成CTCLabelDecode不知道怎么样,还是需要写一个DistillNRTR解码代码

@yang-chenyu104
Copy link
Author

也会出现valid__ratio键值错误

@yang-chenyu104
Copy link
Author

采用MultiScaleDataset读取数据集出现properties为int数据

@yang-chenyu104
Copy link
Author

在配置文件训练那里加入RecResizeImg后又出现网络问题,输入的tensor要求4-D or 5-D,但是收到的维度为3

@yang-chenyu104
Copy link
Author

核查过数据输入的image类型为[96,3,48,320]但进去变成[96,240,256]

@yang-chenyu104
Copy link
Author

目前问题解决,就是之前问题采用蒸馏的配置文件训练,后面改成v4_rec配置文件是正常的,后续将蒸馏配置修改继续提高识别率,谢谢大佬开源工作!!!

@704572066
Copy link

4-D or 5-D

在配置文件训练那里加入RecResizeImg后又出现网络问题,输入的tensor要求4-D or 5-D,但是收到的维度为3

4-D or 5-D 这个问题是怎么解决的

@spencergg
Copy link

目前问题解决,就是之前问题采用蒸馏的配置文件训练,后面改成v4_rec配置文件是正常的,后续将蒸馏配置修改继续提高识别率,谢谢大佬开源工作!!!

请问具体是怎么做的呢,能详细说下么

@anonymous530
Copy link

目前问题解决,就是之前问题采用蒸馏的配置文件训练,后面改成v4_rec配置文件是正常的,后续将蒸馏配置修改继续提高识别率,谢谢大佬开源工作!!!

@yang-chenyu104 你是 v4-distillation 预训练模型 + rec_v4 的 conf,训练的吗?最终训练效果如何呢?

@yang-chenyu104
Copy link
Author

yang-chenyu104 commented Dec 7, 2023 via email

@anonymous530
Copy link

是hg那个文件准确率目前效果是最好的,蒸馏那个需要改配置文件,hg那个路径正确就可以训练

---原始邮件--- 发件人: @.> 发送时间: 2023年12月7日(周四) 晚上9:05 收件人: @.>; 抄送: @.@.>; 主题: Re: [PaddlePaddle/PaddleOCR] 遇见ocrv4训练问题 (Issue #11149) 目前问题解决,就是之前问题采用蒸馏的配置文件训练,后面改成v4_rec配置文件是正常的,后续将蒸馏配置修改继续提高识别率,谢谢大佬开源工作!!! @yang-chenyu104 你是 v4-distillation 预训练模型 + rec_v4 的 conf,训练的吗?最终训练效果如何呢? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

@yang-chenyu104 好的,谢谢

@jtcjump
Copy link

jtcjump commented Feb 29, 2024

遇到同样问题,baidu不升级修补下这个问题,来自于2024.2

@jzhang533 jzhang533 added triaged this issue has been looked, and triaged. training this is a training related issue help wanted this issue needs help labels Apr 10, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
help wanted this issue needs help training this is a training related issue triaged this issue has been looked, and triaged.
Projects
None yet
Development

No branches or pull requests

7 participants