search

uni-medical / SAM-Med3D

SAM-Med3D: An Efficient General-purpose Promptable Segmentation Model for 3D Volumetric Medical Image
Apache License 2.0
508 stars 65 forks source link

Use SAM-Med3D encoder for inference without prompts for automatic segmentation #79

Closed SergioRodLla closed 1 month ago

SergioRodLla commented 2 months ago

Hi, thanks for all the work done!

Is it possible to leverage just the SAM-Med3D encoder (with sam_med3d_turbo) for inference without click prompts? I want to use its encoder with a custom decoder in an automatic segmentation task. My idea is to fine-tune a decoder only, while taking advantage of the features extracted by SAM-Med3D encoder (frozen). Would click prompts still be needed in this case?

Best, Sergio

blueyo0 commented 2 months ago

Hi, good idea! SAM-Med3D has the potential to work as a powerful pre-training for 3D medical automatic segmentation. You can fine-tune the decoder only and discard the prompt encoder.

Here's some reference results for transfer learning: image

SergioRodLla commented 1 month ago

Hi. Thanks for the feedback!

blueyo0 commented 1 month ago

😄 Ok, close this issue now since it is solved.