[Feature] Fine-Grain object search using Grounded SAM or similar

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

I absolutely love the implementation and integration of CLIP in immich, and in generell i think it works pretty well.
Given that a few years in machine learning or deep learning advances feel more like decades, many more interesting approaches have been open sources when it comes to object detection.

I find an interesting approach would be also to have Grounded SAM to localized precisely in image the object of desire.
Or another interesting application case would be that one can select an object in one image and find the same object in all the other images.

This would in generally also help for identifying pets or other objects which can be struggling for CLIP, in particular when it comes to detect small object in the larger image context.

I love to here the opinion on it!

Platform

• Server
• Web
• Mobile