Comment by Randomoneh

I fail to see novelty here. What's the size difference between the model and and all of the 64x32 image training data? If the difference is not significant, you're basically almost just scrubbing a video, right?