How to obtain predicted frame label of a test video?

[Youthfeng123]

Hi there, thanks for your great work in action segmentation, which worths in-depth study of value.
I've download your open source code and successfully run evaluated experiments. In order to simplify the results, I'm trying to visualize the predicted label on input video. Noticing that you've write an instruction about "Action Segmentation", I think I don't have write the code myself entirely. But I don't quite understand what "using ASFormer by the previously extracted frame features" means. Could you explain it briefly for me? Or how to get predicted label of a chosen tested video base on your code?
thanks a lot!

[ttlmh]

Hi, thanks for your interest in our work.

As for the experiments on action segmentation, we mainly used the ViT model trained by Bridge-Prompt framework to extract the frame-wise features for the videos. Then, we took some existing backbones (e.g. ASFormer) to train the action segmentation models based on the extracted features. You can refer to the detailed instructions provided by @yo3nglau in Issue #2.