Baseline: run videollama for yes or no answer to determine whether a given video contains unusual activities, get the baseline numbers for the oops datasets Get the baseline number with VLM
Baseline with vision-only model
For the videos that were detected as unusual, ask for the “why? Question” / “describe why this video is unusual” in a separate exp.
(oops dataset paper has a pre-trained model that can be used as another baseline)