SmilingWolf/wd-tagger · Feature Request: Average of several images

In a situation where someone had multiple images of a style or subject that they wanted analyzed, I believe it could be beneficial to encode them all with the vit and then average their outputs before tags are inferred.

This is assuming that it works something like: Encode image to get a single vector -> Infer tags from the vector. Where the single vector per image can be averaged to reduce outliers.

I'm uncertain if this would be better than inferring tags for each image independently and then averaging the resulting tag scores.