Hi, Thanks for your excellent work. I have two questions.
Have you ever tried removing the mask operation during training and inference, and directly using a weight of 0.5 to fuse the two warping results? I wonder how much impact will this have for the model.
When I infer some hard cases, there are some artifacts in results. Do you have any suggestions or measures for it?
Thanks a lot!
I believe the effect will be poor. I don't remember if I have done any related experiments
I don't know particularly good method, maybe you can calculate the similarity between warped_img0 and warped_img1 to delete some interpolated frame results and directly copy the previous frame
Hi, Thanks for your excellent work. I have two questions.