Why do we input a 16-frame video and reconstruct only 13 frames?

[QChencq]

Thanks for your exciting work!
We input a 16-frame video. When decoding, the first block of the video is reconstructed as 1 frame, and the second block is reconstructed as 12 frames. When concated ​​together, there are only 13 frames. Why is this?

[qqingzheng]

Hi. Because we use Causal Convolution, you should input a video with 1 + t_down * n frames like 9, 17, 25, 33. Here, t_down = 4.