Cannot reproduce performance on forn_2v

dunbar12138 / DSNeRF

Code release for DS-NeRF (Depth-supervised Neural Radiance Fields)

https://www.cs.cmu.edu/~dsnerf/

MIT License

746 stars 126 forks source link

Cannot reproduce performance on forn_2v #80

Closed AliceMegatron closed 1 year ago

AliceMegatron commented 1 year ago

Hi~ Thank you very much for the open source code and paper

I tried to train a model with default config，but I got poor performance on forn_2v.I strictly followed the instructions to start the reproduction experiment，However, my experimental results are quite different from the released results； result of mine： 000 002 007 log of last iter：

Am i missing something ? Really appreciate for your help.

phhyk1995 commented 1 year ago

Hi~ I just meet same problem，I got only 16.98 PSNR on forn_2v with release config，Do you have any idea now？

AliceMegatron commented 1 year ago

Not yet. Due to poor performace of my rendered result, I believe I miss something indeed,So I start working with other baselines.

dunbar12138 commented 1 year ago

Hi,

A test PSNR of ~19 is expected on the horns data. I would suggest several things to check:

Are the latest code and data being used?
Are some parameters being changed? (e.g., number of sampled rays to fit into memory)
Check the released pre-trained model and see whether the results are similar.

Thanks for your interest!