search

bowang-lab / MedSAM

Segment Anything in Medical Images
https://www.nature.com/articles/s41467-024-44824-z
Apache License 2.0
2.84k stars 385 forks source link

LiteMedSAM without prompt #316

Closed Yiming0131 closed 1 week ago

Yiming0131 commented 2 weeks ago

I am currently fine-tuning and testing LiteMedSAM without providing box prompts during the process. Interestingly, the test results show that it still performs well in tumor detection and segmentation. Could this suggest that it has the potential to be used as a fully automated segmentation network? Looking forward to your early reply

JunMa11 commented 2 weeks ago

Hi @Yiming0131 ,

SAM like model is designed for promptable segmentation. Although it can be modified to auto-seg model, the traditional segmentation network is a better choice for fully automatic segmentation tasks, such as u-net, segformer et al.

Yiming0131 commented 1 week ago

Thank you for your response! I have one more question: If no box prompt is provided, how does the network handle image segmentation? Does it treat the entire image region as the box prompt? Or will there be other ways to auto-generate prompts？

JunMa11 commented 1 week ago

hi @Yiming0131 ,

SAM is a promptable segmentation model. box prompt is necessary!