I would like to train a model to classify frames in a video (and identify "best" frame for something I want to locate, according to my training data).

Is SAM-2 useful to use as a base model to finetune a classifier layer on? Or are there better options today?