retsuh-bqw / SRFormer-Text-Det

[AAAI 2024] SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

precision、recall、hmean的结果都是0?

Cookie477 opened this issue · comments

作者您好,我用您预训练的模型评估ctw1500数据集,请问为什么precision、recall、hmean的结果都是0?

作者您好,我用您预训练的模型评估ctw1500数据集,请问为什么precision、recall、hmean的结果都是0?

The pre-trained weights we provided is one of model variant with only 2 Segmentation&Regression Chunks.
You should correspondingly change the config file to load the weights correctly. And you can change the configuration in the line I quoted below:
https://github.com/retsuh-bqw/SRFormer-Text-Det/blob/c71354939e576353f5ae202f2ca559ad3fd255e6/configs/SRFormer/Base.yaml#L26C8-L26C8

作者您好,按您的要求更改配置后,仍然出现这个问题;之前也用微调模型进行评估过,也是同样的问题。
检查运行过程,发现出现以下问题。麻烦作者帮忙看一下,是不是因为这个原因,该如何解决?还是其他版本问题?万分感谢!
image
结果:
2

I retested the uploaded models and they all worked correctly and reproduced the results:
截屏2023-12-15 00 35 06
The warning info you noted may not be a problem. Are you using the dataset we provided?