Hi author, thanks for your great work. I would like to ask if it is possible to perform autoregressive rollout to generate long videos using the bidirectional teacher model with sliding window inference.
As in the paper, you seem to train the student model simply using 10s videos without autoregressively generating longer videos, and sliding window is used during inference to generate long videos in an autoregressive way. So I suppose this autoregressive generation method should also be applicable to the teacher model (although results may not be with good quality).
Please tell me if I understand wrongly. Thank you a lot in advance.
Hi author, thanks for your great work. I would like to ask if it is possible to perform autoregressive rollout to generate long videos using the bidirectional teacher model with sliding window inference.
As in the paper, you seem to train the student model simply using 10s videos without autoregressively generating longer videos, and sliding window is used during inference to generate long videos in an autoregressive way. So I suppose this autoregressive generation method should also be applicable to the teacher model (although results may not be with good quality).
Please tell me if I understand wrongly. Thank you a lot in advance.