search

SysCV / sam-hq

Segment Anything in High Quality [NeurIPS 2023]
https://arxiv.org/abs/2306.01567
Apache License 2.0
3.73k stars 224 forks source link

When I use ''multimask_output=True'', only one mask is generated #62

Open sssmallmonster opened 1 year ago

sssmallmonster commented 1 year ago

Thank you for your efforts. I've tested SAM-hq; however, I encountered issues while running the model. Specifically, when utilizing multi-prompts along with the previous masks, I observed that only a single output mask is generated. Could you please provide some insight into this matter?

Furthermore, during my testing, I employed the sam_hq_vit_h.pth checkpoint. Although the accuracy of the obtained masks exhibited some improvement compared to the results from SAM, they still fall notably short of the quality demonstrated in the official SAM website's demo. Your guidance on these matters would be greatly appreciated. 屏幕截图 2023-08-16 154619 屏幕截图 2023-08-16 153808

ymq2017 commented 1 year ago

Hi, we select the mask with the highest score for multi-mask output. The corresponding code is in this line. It is also used in COCO evaluation or other quantitative evaluations.

ukaprch commented 12 months ago

MultiMask=True by definition as used in regular SAM produces the 3 highest scored images. So why can't HQ do the same?