AJ Piergiovanni

This has to do with the lines that select the features to send back to the input space (roughly line 68 in the demo code): feat = net.blobs['pool5'].data feat[0][feat[0] <...

I think this would be open to interpretation of the idea and results. I see no reason why this cannot be used to for multiple features. For example, if you...

You'll have to change the model defined in invnet.prototxt to start at the correct layer (for example, pool4 instead of pool 5), which should be as simple as deleting a...

That import isn't needed here, it was from some other experiments and I forgot to remove it. The file has been updated, removing that dependency.

I'm not sure what the RCE repo is, though I am happy to help with running the code available in this repo. Regarding the training details: The learning rate was...

Those first errors are because the models are from a different version of pytorch. They won't change anything as long as you are using pytorch 0.3.x. For the second error,...

Due to the strides and max-pooling in I3D, it temporally downsamples the video by a factor of 8. So if you input 64 frames, you'll get 8 (temporal) output predictions....

Based on the implementation details in that paper, the only normalization done is to rescale everything to [-1,1]. I do handle that in https://github.com/piergiaj/pytorch-i3d/blob/05783d11f9632b25fe3d50395a9c9bb51f848d6d/charades_dataset.py#L37 and https://github.com/piergiaj/pytorch-i3d/blob/05783d11f9632b25fe3d50395a9c9bb51f848d6d/charades_dataset.py#L54-L55 I'm not sure if...

Yes, for flow I know this version is correct because the model is fine-tuned with this normalization. For the RGB, it isn't clear. It isn't mentioned in the paper or...

These lines: https://github.com/piergiaj/pytorch-i3d/blob/master/train_i3d.py#L115-L119 Only apply the gradient step when in training model. Combined with https://github.com/piergiaj/pytorch-i3d/blob/master/train_i3d.py#L86 the gradients from the validation step are never applied. For efficiency, the loss.backward() could be...