about "Feature-Crop baseline"

[nhw649]

Hello, I would like to ask how the "Feature-Crop baseline" mentioned in the paper crops features using a mask? Is there any specific paper that I can refer to?

[HarborYuan]

Hi @nhw649 ,

Thanks for your question.

As mentioned in the paper, the "feature-crop baseline" is quite simple. It first gets the object mask with SAM and then extracts the instance feature from the CLIP backbone with mask_pool.