PaddlePaddle / PaddlePaddle/Paddle

PaddleOCR在训练DBnet_r50_vd_db的时候,无法复现精度

Open
#54,589 4 comments 0 reactions 1 assignee View on GitHub

@YuanRisheng is already working on this.

Since Jun 13, 2023.

status/new-issue type/docs
Dominant language
C++
Stars
24.1k
Forks
6k
Avg merge
4d 17h
Merged PRs (30d)
60

Description

文档链接&描述 Document Links & Description

训练环境:GPU A100,paddlepaddle 2.4
操作系统:ubuntu 18
paddleocr版本:2.6
问题描述:实际跑出来的三个精度指标无法到达官方标称的结果,config链接的配置和提供的日志文件的config也有差异
通过增加eval次数也无法达到精度结果;

使用配置文件:configs/det/det_r50_vd_db.yml,
连接https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/configs/det/det_r50_vd_db.yml
启动训练命令:
python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7' tools/train.py -c configs/det/det_r50_vd_db.yml

实测GPU精度结果:
[2023/06/05 09:50:35] ppocr INFO: best metric, hmean: 0.7859162169108198, is_float16: False, precision: 0.8428886438809261, recall: 0.7361579200770342, fps: 33.370472597937706, best_epoch: 1000
官方给出的日志精度结果:
[2020/11/26 20:53:46] root INFO: best metirc, recall: 0.7871930669234473, best_epoch: 1062, fps: 24.92656319259578, precision: 0.8641649048625792, hmean: 0.8238851095993953

官方给出的config:
https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/configs/det/det_r50_vd_db.yml
官方给出的日志配置:
https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar
其中主要差异在预训练文件和backbone

请提出你的建议 Please give your suggestion

复现训练精度结果,更新推荐的配置文件

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.