Problems with multi-cat detection

#2
by CapitanMurloc - opened

If you test with two cats, the mask fails. Is there any extra adjustment to be made to detect all possible cats in the photo?

felinos.jpg

Hi, thanks for your interest. Our current models personalize SAM by a certain object, e.g., a black cat or an orange cat, not a semantic category. If you need to segment both of them, you should give a one-shot image containing both two cats, or in respective two-shot images. Then, you can store two target embeddings to respectively segment them. We will further extend PerSAM to be personalized by a semantic category, e.g., cat, to segment all cats in test images.

justin-zk changed discussion status to closed

Sign up or log in to comment