PaddlePaddle / PaddleOCR

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
Apache License 2.0
38.98k stars 7.31k forks source link

V4识别蒸馏模型预训练权重以及训练出现KeyError: 'valid_ratio' #11988

Closed marsbzp closed 2 weeks ago

marsbzp commented 3 weeks ago

V4识别蒸馏模型预训练权重以及训练出现KeyError: 'valid_ratio',你们有做过验证吗,V4问题一大堆,预训练权重和yaml对不起来的ch_PP-OCRv4_rec_distill.yml 它的训练权重在哪

UserWangZz commented 3 weeks ago

抱歉,我找相关同学确定一下问题

TingquanGao commented 3 weeks ago

可以提供一下训练命令吗?我们这边复现一下。

marsbzp commented 3 weeks ago

ch_PP-OCRv4_rec_distill.yml 用这yaml训就复现啊,这issue之前也有人提过你们一直不修复的吗

marsbzp commented 3 weeks ago

预训练权重能给个匹配的吗

tink2123 commented 2 weeks ago

v4 有一些遗留的问题,不推荐使用蒸馏的配置。bug在排期修复中,建议使用 https://github.com/PaddlePaddle/PaddleOCR/blob/main/configs/rec/PP-OCRv4/ch_PP-OCRv4_rec.yml 来训练