[RL Unplugged] Pre-Processing of Atari Datasets

[arjung128]

Was max-pooling of the last two frames performed in the pre-processing of the Atari datasets?

The DQN Replay Dataset (An Optimistic Perspective on Offline Reinforcement Learning (ICML, 2020)) does max-pooling of the last two frames, and as a result, the images seem to be rather blurry, while the images in the RL Unplugged Atari dataset seem a lot clearer.