nex-mpi / nex-code

Code release for NeX: Real-time View Synthesis with Neural Basis Expansion
MIT License
594 stars 73 forks source link

Render depth #28

Closed shanemankiw closed 3 years ago

shanemankiw commented 3 years ago

First of all, thanks for sharing such a great work.

I noticed that although there is an option to 'render_depth' in render_video, it seems that depth rendering is not implemented?

pureexe commented 3 years ago

render_depth is used to render the pseudo depth map to compare with Deepview which looks like this.

https://youtu.be/UD7I9gXIVsc?t=267

image

It will replace the output color rgb from MPI with the depth instead.

Here is its implementation. https://github.com/nex-mpi/nex-code/blob/eeff38c712ac9a665f09d7c2a3fdf48ae83f4693/train.py#L361-L370

However, our depth is quite bad. since it try to represent some reflection (for example in the reflection on the CD) into the plane. So, the pseudo depth map looks wrong. We haven't shown the depth result on the main paper but feel free to compare with it.

shanemankiw commented 3 years ago

Oh that's it, I oversaw that. Thank you very much for the reply!

TianYinchen56 commented 3 months ago

Thanks to the author's answer, but I still have a question, I added -render_depth when I trained the data as well as when I rendered the data. I checked that render_depth was turned on, but the rendered pictures and videos are still ordinary pictures rather than depth pictures, hoping to get an answer from the author, thanks again