Checkpoint for each dataset

[ByteLemon]

Thanks for your wonderful work. Would you be able to share the trained checkpoints for each dataset?

[vimar-gu]

Thanks for the interest in this work. However, I currently don't have sufficient cloud space to upload these checkpoints.

The fine-tuning process is quite straightfoward and fast. It takes less than 1h in a 4090 GPU. For larger datasets, you need to separate the datasets into multiple 20-class subsets, and fine-tune an individual model on the subset. Tbh, our lab's GPU resource is quite limited, which restricted the fine-tuning parameter setting. If you have sufficient resource, by tuning the hyper-parameters it can easily achieve better results.