Using equirectangural video as input for the pipeline

[Inshu32]

I have some 360 degree videos which I would like to convert into point clouds of decent quality. I went through your docs but couldn't find any specific guideline on how to use the framework for videos. Can you please help. Also, do I need to create some training data to use FreDSNet for my custom data ?

[Sbrunoberenguel]

Hello,
The network is not meant for videos, only single images. You could go frame by frame but I cannot guarantee the consistency of the output data between consecutive frames.
FreDSNet is not trained for zero-shot inference, so there would be some domain gap with your custom data (most likely the results will be worse than those reported in our article).
Hope this helps!