Video embedding does not match the model

[jameslovestudy]

Thanks a lot for the open source. I'm new to deep learning and I'm having problems running the model after processing the dataset. Using visual_frontend to extract the video embedding and store it as npz, the final dimension I get is n×512, while the dimension of ImageNet input is required to be 256, may I know how to solve this problem without affecting the training result.

[zexupan]

Hi, in the model, you can change
self.v_ds = nn.Linear(256, 256, bias=False)
to
self.v_ds = nn.Linear(512, 256, bias=False)

[jameslovestudy]

Thanks for the reply. I change N which is the parameter of the video decoder to 512 as well and it works fine.