coltonstearns / dynamic-gaussian-marbles

MIT License
112 stars 9 forks source link

Render training back to video #10

Open heimrabe opened 1 week ago

heimrabe commented 1 week ago

Hi thank you very much for your work. I am wondering, is it possible to render the training back to the resolution and size of the original video? I see that the rendering is done with a nerfstudio function which adds the tracks on the reconstruction. I am trying to train on some custom real world videos and would like to get a video which is a reconstruction of the real video from the training.

coltonstearns commented 4 days ago

Yes, just comment out line 36 of src/visualization/render.py (i.e. the line with cameras.metadata['viz_tracking'] = torch.ones_like(cameras.times) * 0). This will turn off the point-track rendering. I can also add this toggle as an option in a future code push. Let me know if you have any more questions about this!