cyhsu14 / omninerf Goto Github PK

I have a question regarding the coordinate transformation section of your code. I noticed that only the origin coordinates (for translation transformation) are provided. What should I do if my data involves rotation changes?

Hi @cyhsu14 , may I know how we can generate the depth maps for the training data? (the ones that we generate in Generate_data.ipynb)

Hi @cyhsu14 @sangminkim-99

Shouldn't we add the following line before 630 in run_nerf.py?

            rays_rgb_test = torch.reshape(rays_rgb_test, (11, 512, 1024, 3))

Majorly because the dimensions of rays_rgb_test and rgbs would not match in the either case

In the following lines of run_nerf.py:

        if args.use_depth:
            depth_loss = torch.abs(dep - target_d).mean()

isn't dep a rendered depth map and target_d from the direction map? Could you please clarify this?

Hi,

Can you provide the code for generating perspective views with rotation like this?
Seems like the current source code only generates 360 sequences with translation only.
Thanks!

What image distance would be good to create a good model ?
Can we extract the mesh ?