Project:
https://github.com/Hritikbansal/videophy/tree/main/VIDEOPHY2
caption: original prompt in the dataset
video_url: generated video (using original prompt or upsampled caption, depending on the video model)
sa: semantic adherence score (1-5) from human evaluation
pc: physical commonsense score (1-5) from human evaluation
joint: computed as sa >= 4, pc >= 4
physics_rules_followed: list of physics rules followed in the video as judged by human annotators (1)
physics_rules_unfollowed: list… See the full description on the dataset page:
https://huggingface.co/datasets/videophysics/videophy2_test.