search

dk-kim / CAFE_codebase

Official PyTorch implementation of "Towards More Practical Group Activity Detection: A New Benchmark and Model"
9 stars 1 forks source link

How to test the custom video? #3

Open JihuaPeng opened 1 month ago

JihuaPeng commented 1 month ago

This is an excellent work for group activity detection. May i know how to test the custom video? Can you release this part of the code?

dk-kim commented 1 month ago

I'm not sure what you mean "custom video". The evaluation code for Café test-set has already been made public, so are you referring to the raw video?

JihuaPeng commented 1 month ago

I'm referring to the videos in the wild, such as multi-player videos downloaded from youtube. May i use the evaluation code for Café test-set to inference the videos in the wild?

dk-kim commented 1 month ago

I see. I think it would be difficult to directly apply the method to videos in the wild without additional training. First, group activity detection (GAD) task in our paper assumes that the bounding boxes in the input videos are provided by ground-truth or an off-the-shelf tracker. Second, the method is designed to deal with predefined group activity classes, so it cannot detect group activities outside of these class categories.