Open wjmsxbd opened 1 week ago
During model training, we employed a frame sampling operation to ensure the motion amplitude of the generated videos, ultimately using data at a frequency of 4 Hz for training. As demonstrated on our project homepage, the generated short videos are 2 seconds long, which corresponds to 8 frames at 4 Hz.
Thanks for your awesome work! In your paper, you mentioned that the data obtained was 12Hz, but during training, the input data was also 12Hz or 4Hz, because in the later implementation details, it was mentioned that the FVD was calculated at 4Hz. Looking forward to your reply!