Asking for dataset to know if the high loss is because of bug or difference in dataset

[InvBoiz]

Dear authors,

After reading closed issues, i saw that there are a small amout of people got very bad result after training their model, and you said it maybe because of training data. Can you share your dataset so we can try to improve our model.

I'm sorry for the inconvenience that i caused.
Looking forward to hearing back from you.

[crsl4]

Thanks for your message!
I am not sure we are able to share all the training data, but we might be able to share the trained model.
@YurenSUN can we do this?

[YurenSUN]

Thank you for reaching out! We actually do not have a trained model for all the classes as in our experiment, we randomly selected classes and applied our training and testing pipeline to them instead of focusing on specific classes.

I think that is also why we put up a sample training dataset and colab file to go through the training and testing process. My thought is that maybe you could try using our sample dataset provided in this readme and to see whether this might work with your codes. This is also the dataset that was used in the previous issues. Would this make sense or help?