CLIP
CLIP copied to clipboard
Has anyone run experiments on actionrecognition datasets ?
I saw that the experiment was done on UCF101, but according to the idea of the paper, text information needs to be provided, I don't know how to get text.
simple put the ground truth label as the text.
Have you seen the list of classes and templates here? https://github.com/openai/CLIP/blob/main/data/prompts.md#ucf101