search

bmild / nerf

Code release for NeRF (Neural Radiance Fields)
http://tancik.com/nerf
MIT License
9.83k stars 1.37k forks source link

Results of applying NeRF #146

Open hyunJIN7 opened 2 years ago

hyunJIN7 commented 2 years ago

Hi, First of all, I am a big fan of this project, and I really appreciate opening your source code publicly.

I applied NeRF to my datasets that was obtained through the ios_logger app. This is data loaging code.

The following is the result of training 200,000 times. [Origin Image] Origin Image

[Result RGB image] Result RGB image

[Result Depth image] Result Depth image

As you can see from the images, the result is not good. The RGB image is very blurry and the depth image did not extract properly.

  1. Have you ever applied data by acquiring it with an iPhone?
  2. Do you know why the result came out like this? Do you know how to slove this problem? What should I do to get proper result from the data I acquired?

Thank you in advance for reading this long post!

deepankverma commented 2 years ago

Maybe it has to do with the number of photos and probably a reflection from the screen.

qbeer commented 1 year ago

Have you fixed this? I think your poses are incorrect. Because the rendered image is from a different view than the original image. You might have missed that.

I can't really understand what you are doing here https://github.com/hyunJIN7/nerf-pytorch/blob/3f30bca372aaff58264c9f04f6932ea78a450839/load_arkit.py#L22 exaxtly. It seems you are overwriting your poses kind of with the NeRF-Synthetic poses.