mxin262

@whitebabycat Use this [link](https://drive.google.com/file/d/1tC_DIRuiN8vGAD5U00z2VF4IwKtL171l/view). Baidu Web disk in the readme is recommended.

@jeong-tae Refer to this #47

You are rights. Limiting the number of ground truth is required. You can try to fix the code [here](https://github.com/mxin262/SwinTextSpotter/blob/f2ebbe87810a14a28f59b91c2f6c318de461ec8d/projects/SWINTS/swints/swints.py#L222). I will update it later.

Refer to this #47

You can download the annotation file [here](https://drive.google.com/file/d/1Yx3GRRUogjYYDUrprexGnPerFX3H6r_3/view). There is a problem with the original path. I update it. "rec" means the text.

Actually, there have 1255 imgs which is consistent with the official version. You can find in detectron2 log [here](https://user-images.githubusercontent.com/62656382/183273903-294f99f9-0cf5-423a-9c4c-e610079e8700.png)

Hi, If you want to train custom dataset, you can follow this [document](https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html) in detectron2.

You can reduce the number of query or input image size to accelerate.

You can build directly by running python setup.py build develop. Don't not need to run the command %cd detectron2 !python setup.py build develop. I update the readme.

能否提供更多信息