Véronique Prinet - Motion Selective Prediction for Video Frame Synthesis

Motion Selective Prediction for Video Frame Synthesis

Bird sequence


Experimental protocol: trained on 50 frames (1.5 period), prediction on 25 frames. Context of four frames. Frame size: 256x256.


Training (12fps) Prediction - Ours-FDN (7fps)
Comparisons (B1, ours-M1, ours-FDN) (2fps)

Selector activation curves (motion periodicity: 30 frames)