Thanks for your interesting work. I've checked the code and found that in the temporal attention implementation, each time the newest noisy frame will condition on previous almost clean frames for denoising. I'm wondering why in this case the auto-regressively generated frames will not face quality degradation. Could you provide some insights about this? Thanks.
Thanks for your interesting work. I've checked the code and found that in the temporal attention implementation, each time the newest noisy frame will condition on previous almost clean frames for denoising. I'm wondering why in this case the auto-regressively generated frames will not face quality degradation. Could you provide some insights about this? Thanks.