search

vye16 / shape-of-motion

MIT License
798 stars 56 forks source link

Rendering reconstruction back to video #35

Open heimrabe opened 2 months ago

heimrabe commented 2 months ago

Hi, thank you again for the amazing work. I was wondering how to render the generated 4D representation back to a monocular video. I would like to have a video of the rendered scene from the same point of view as the original monocular video used. I tried using run_video.py, but the output mp4 video is always empty.

qianqianwang68 commented 2 months ago

Hi, this function should save rendered videos from the original videos' viewpoint. Does that suffice?

heimrabe commented 2 months ago

Hi, thank you for your reply. If I understand the code correctly the videos are saved only for the Iphone dataset. I am currently training on a custom dataset and would need the videos for those. So far, I have modified the run_video.py code, I am using the FixedTrajectoryConfig and using the train_w2cs to render the videos. But I am not sure whether that's an appropriate solution or not. If I want to render the reconstruction for videos of a custom dataset back to a video from the original videos' viewpoint, how should I proceed?

heimrabe commented 1 week ago

Hi @qianqianwang68, could you let me know whether that make sense as an approach? I would need to create videos of the reconstructions after the training is done on various monocular videos