xiaohangzhan/conditional-motion-propagation

Code for our CVPR 2019 work.

conditional-motion-propagation's Issues

Is it a self-supervised method?

Thanks for your interesting work.
I have a small question. The paper title implies your method is self-supervised, but I notice your network is supervised by ground truth which is generated by another optical flow network.
So is it proper to be called as a self-supervised method?

Was the pre-training model corrupted？

Awesome work

I do not have any questions, just want to say this work is aaaaaaaawesome.
It deserves more attention from the self-supervised learning community.
I believe the conditional motion propagation pretext task will play an important role in the future, serving as one crucial component of a multi-task self-supervised learning framework.

Do you have scripts for computing flow for my own dataset?

I want to train on my video dataset

Can you please release the tutorial of eval？

Can you please release the tutorial of eval？
including the data structure.
it will help a lot !

video generation ipynb demo

Thanks for the great contribution. I was wondering if you are able to provide the demo code for the video generation as it was shown in the youtube video. Preferably, the code generates a .gif output, given the arrows/anchors.

Thanks,

How did you get the initial guidance here?

The article says that the guidance is sampled from the target flow, but what is the basis of the "sparse motion+mask" we input at the beginning? At this time, the static image does not have a corresponding optical flow diagram.