MTC-VAE logo MTC-VAE

For c ≥ 5, an extremely poor reconstruction quality was evidenced. This may be due to the overall length of the videos of this dataset, which go from 6 to 13 frames, so interaction between chunks is not possible to attain when chunks are too long. Perceptually the best reconstruction is evidenced for c = 1, although c = 3 also presents some identity preservation and correct motion rendering from the driving video.