hzhuangdy

Cudnn error in validation step

1

Hi, I wonder why this happened after a few iters of validation? 

Visualization of datasets

1

Is there any way to visualize my training datasets? For example, when I train my own datasets with DBNet, is there any way to visualize the cropped image/shrink/threshold map? Or...

You guys do a fascinating job in the document analysis field. I have a question about the preprocess of the document image in Dit. Why would you use the mean...

同时开启AMP和单机多卡训练DBnet++，loss经常Nan

2

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：

[Bug] Slow training-time evaluation

16

### Prerequisite - [X] I have searched [Issues](https://github.com/open-mmlab/mmocr/issues) and [Discussions](https://github.com/open-mmlab/mmocr/discussions) but cannot get the expected help. - [X] The bug has not been fixed in the [latest version (0.x)](https://github.com/open-mmlab/mmocr) or...

Switch pretrained Decoder(for example, uses RoBERTa-xlm to replace mBART)

2

You guys did a great job in VDU field. Congratulations! By the way, I wonder if I can replace the mBART by RoBERTa-XLM on finetune process without doing pretraining jobs...