Hi, thanks for the wonderful work!
There is one thing that I realized after some experiments. The accuracy of PerSAM also depends on the reference image chosen. Choosing different images gave different results for the PerSeg dataset.
I overcame it by using the image which is closest to the mean of all the feature maps given by SAM's encoder.
Hi, thanks for the wonderful work!
There is one thing that I realized after some experiments. The accuracy of PerSAM also depends on the reference image chosen. Choosing different images gave different results for the PerSeg dataset.
I overcame it by using the image which is closest to the mean of all the feature maps given by SAM's encoder.