Video completions on train set vs test set

Arrays on the left are completions of training videos, and arrays on the right are completions of test videos. In each 5x4 block of videos, the top row shows train/test videos, and the remaining rows show three FDM completions, conditioned on the first 36 frames. Observed frames are shown with a red border, and we mark the end of the video with a checkerboard pattern.

MineRL (sampled with Hierarchy-2)

GQN-Mazes (sampled with Long-range)