Expected size of video

[alaflaquiere]

Hi,

Your videos looks great, congrats!

I just tried running your Colab demo with (W=540, H=960) video, but I encountered the following error when running. prediction = processor.step(frame_torch, mask_torch[1:]):

RuntimeError: Sizes of tensors must match except in dimension 2. Expected size 480 but got size 960 for tensor number 1 in the list.

Does the model expect a particular video size?
If that's not the case, any idea what the issue might be?

Thanks!

[hkchengrex]

Hi,

It should work with any size (we did most of the experiment in 480p though). Can you check if the frame and the mask have the same size?

[alaflaquiere]

Actually, going back to it today, I realized that the video loaded in the mp4 variable is not the one fed to the network. Instead, it is loaded again in the cap variable, which I hadn't changed.
So totally my fault! It's working now. :)

Thank for the quick reply.

[hkchengrex]

Ah, my bad. I literally copy-pasted that visualization code from stackoverflow without realizing that it creates the mp4 variable. I have made that inline to prevent future confusion. Thanks!