Few-shot fine-tuning

I am curious to know how much data that we must need for few-shot finetuning.

I would be a great help if I could get the right parameter settings for fine-tuning the model with that small amount of dataset.

Finally, a clue to make custom dataset compatible with fine-tuning the SAM3 model.

Thank you in advance!

Start by reading the repository's fine-tuning entry points and example notebooks to identify the documented training configuration and dataset format. The work is complete when the few-shot data requirements, relevant parameter settings, and custom-dataset compatibility guidance are clearly documented for SAM3 users.

Written by the indexing model from the issue text.