About frames number of dataset about movinet-pytorch HOT 4 CLOSED

It's generally ok to train on another dataset with different frames. You don't have to input the same number of frames of the original paper.

from movinet-pytorch.

Thanks for your reply.

And there is another confusion. In the paper, "Similar to training, we use the stream buffer to cache activations between subclips. However, we can set the subclip length to a single frame (Tclip = 1) for maximum memory savings". Is it means we can train with Tclip = 8 and inference with Tclip = 1, training frames number is different with inference frames number?

If we want to inference with Tclip = 1, we must train with Tclip = 1?

from movinet-pytorch.

You can train with Tclips=8 and inference with Tclip = 1, whitout any change in accuracy.

from movinet-pytorch.

You can train with Tclips=8 and inference with Tclip = 1, whitout any change in accuracy.

thanks

from movinet-pytorch.