caoyunkang / Segment-Any-Anomaly

Official implementation of "Segment Any Anomaly without Training via Hybrid Prompt Regularization (SAA+)".
689 stars 73 forks source link

use for image similarity task? #31

Open hnyll opened 6 months ago

hnyll commented 6 months ago

how can i use this to detect anomalies between two images?

like input two images and then detect anomalies for image similarity :O

caoyunkang commented 6 months ago

@hnyll Hi, do you mean that use an existing normal image as prompt/reference? Currently, SAA doesn't support such few-shot mode, but it may be reasonable to calculate the similarity score using PatchCore. Then the similarity score can be used to guide the segmentation process.

hnyll commented 6 months ago

@caoyunkang oh wow i didnt know that you would reply so fast.

I'm still kind of new to things related to this field so please correct me where I am wrong. similarity scores from PatchCore are scores for each patch correct? and these similarity scores is used to segment the parts that are different from the reference and the query?

could you clarify on what you mean by similarity score?