search

UX-Decoder / Semantic-SAM

[ECCV 2024] Official implementation of the paper "Semantic-SAM: Segment and Recognize Anything at Any Granularity"
2.39k stars 118 forks source link

result is not convincible #86

Open sipie800 opened 9 months ago

sipie800 commented 9 months ago

2024-02-06_180916 This is the predicted mask of the demo swint. I used it to test SAM and it just fails. A damn simple task isn't it? Don't think it produces convinciable result. Can we make the SAM-like model a truely robust one?

FengLi-ust commented 9 months ago

I think the first two outputs are reasonable, and our SwinL model gives more robust results. What is the result of original SAM model?