How can I preprocess video data while preserving all frames?

I want to batch video data for action counting task. say, I want to get (Batch, Channel, Frame, Height, Width)
But video data frames are not same each other and It looks like not to possible to apply frame subsampling… because action counting task needs to fed all video frames!

Is there better way to batch video data preserving video frames??

To best of my knowledge, it is impossible.
However, if you construct them into a graph, please see PyG Documentation — pytorch_geometric 2.0.4 documentation

Thank you!! :slight_smile: I’ll try this!