idiap / geonerf Goto Github PK

Hi, thanks for your amazing job. I would like to know when the code will be realsed. Thank you!

hello! Thanks for your excellent work!
I would like to ask how long it will take to train the generalization model on the GeForce RTX 3090 GPUs

You mentioned in the paper that, "The evaluations of the baselines on the DTU MVS dataset [23] are borrowed from the paper MVSNeRF [6]". Is this the case for LLFF and NeRF synthetic as well? I have not been able to reproduce results mentioned in MVSNeRF paper, wanted to confirm if that was the case for you as well.

Loading dtu datasets seems to be a little different from MVSNeRF, why is that?

Hi, @Kanma
This is amazing work. When will the code be released？

hello! Thanks for your excellent work!

My project need a series of new images. But neither fine-tune nor eval can generate more new views or a vedio, which is just like pixelNeRF or nerf-pytorch.

I noticed a file called media in the log file, but I have not ever see a non-empty file called media. Can I get a vedio or more views by this project or modify it?

besides issue #9, I wonder how long the inference of one out-of-distribution scene will take

thank you for your work!
I do not understand why you mask the point which d upper than D, Volume rendering could deal with this kind of points.could you give me some suggestion?tks!

Hi, thanks for publishing your code for the excellent work!

When I was using the code to evaluate the model, I found myself come across "CUDA out of memory" even though I have cut the "chunk " in configuration to a very small number (64). I'm using 4 GeForce RTX 2080 GPUs, and I want to know how much GPU memory is needed for model evaluation.

By the way, I found that maybe it is only able to use one GPU, since no matter how many GPUs I assign in the CUDA_VISIBLE_DEVICES, memory from only one GPU is taken up, then the process is killed.

Best regards,
Shuhong

Hi @Kanma ,
Thanks for your great work. Will the code be released in recent days?

thanks for your geonerf! In training, after geo_reasoner, step in get_rays_pts, why only support bs=1?