search

nianticlabs / diffusionerf

[CVPR 2023] DiffusioNeRF: Regularizing Neural Radiance Fields with Denoising Diffusion Models
MIT License
291 stars 16 forks source link

Reproduce DTU results #12

Closed ChrisWu1997 closed 11 months ago

ChrisWu1997 commented 11 months ago

Hi there,

Thanks for your excellent work. Do you mind sharing the DTU results somewhere or the shell script that can reproduce DTU results? I'm trying to reproduce the results on DTU but not sure if I have the correct setting.

daniyar-niantic commented 11 months ago

Hi. Yes, we can share the DTU results, could you tell us which images you want, i.e. 3-view, 6-view, 9-view, the scene and image name?

ChrisWu1997 commented 11 months ago

Thanks for replying!

Could you share all the results? That is, all the testing frames for 3/6/9 views on all the test scenes. Similar to what FreeNeRF shared here.

jamiewynn commented 11 months ago

Hi Chris,

Unfortunately we no longer have the original DTU few-view frames from the final run, but I did dig up some outputs from our parameter tuning that are almost the same as the final run. The only difference is that the patch regularisation weight in this run is very slightly lower. You can find it here. For each scene, the validation folder contains the images, as well as depth maps. I hope this will still be helpful to you.

Best, Jamie