Training procedure

[BaptistePodvin]

Hi,
Do you use an iterative prompting strategy during the training (Calculating the error region of the first prediction and go back to the mask decoder with an additional point) ?

[blueyo0]

Hi, our training pipeline automatically generates all prompts based on the mask, just like many previous interactive seg methods like SAM and RITM.

BTW, you can refer to our paper https://arxiv.org/abs/2310.15161 for details 😄.