search

computational-cell-analytics / micro-sam

Segment Anything for Microscopy
https://computational-cell-analytics.github.io/micro-sam/
MIT License
365 stars 47 forks source link

use label from detect model to prompt SAM #194

Closed aiyou9 closed 1 year ago

aiyou9 commented 1 year ago

Yolo NAS with SAM https://github.com/RATHOD-SHUBHAM/SegmentAnything

how_to_use_yolo8v_with_sam https://github.com/RATHOD-SHUBHAM/SegmentAnything/blob/master/how_to_use_yolo8v_with_sam.ipynb

constantinpape commented 1 year ago

Hi @aiyou9, can you expand on this issue? Are you proposing integration of object detection for box prompt generation? Please note that this repository is focused on application biomedical imaging.

aiyou9 commented 1 year ago

The challenges in using SAM in this project： https://github.com/XiaodanXing/SAM4organoid#method-description

in the impoverment of SAM, I think the box label from object detection may help to run SAM in 3 terms

  1. prompt for SAM
  2. seg a ROI sub-area to avoid surrounding similar image interference

  3. From the surrounding boxes of a certain box, compare the overlapping parts, and automatically find the position of the negative marker point to prompt SAM

    if this repository is focused on application biomedical imaging, this issue may be not suitable to go on,

mostly zero-shot segment is base on fine-tune SAM, some good work based on SAM

https://github.com/yatengLG/ISAT_with_segment_anything/blob/master/docs/features%20description.md#3contour-mode

so I close this issue