For anyone who can't get the code to run, I made a walkthrough.

[kefeiqian]

Hope it comes in handy.

https://github.com/KefeiQian/BiRSwinT/blob/main/get_started.md

[ktiays]

Your getting started doc are mostly good to me. But we noticed that the pre-trained model file (ckpt_epoch_44.pth) you use is not correct and we have uploaded the right one to Hugging Face. Would you mind opening a pull request to add the doc? Thank you!

[kefeiqian]

Of course! I will change the ckpt_epoch_44.pth and make a pull request today. Thank you so much!

[323417]

@kefeiqian

After I clone the code, put the dataset and 2 pretrained parameters files, run the run.sh (BiRSwinT-maintrain_valid.py) when the output folder is empty.

The original author you gave the accuracy in the training process is slowly increasing, but the first round of accuracy I run is very high (73.4 % as shown in the figure), I don't know if it is based on the training data.

[kefeiqian]

Sorry for the late reply. You can add me through WeChat. My username is as same as my GitHub name(all lowercase). @323417