Does larger MLP affects the final results?

Brummi / BehindTheScenes

Official implementation of the paper: Behind the Scenes: Density Fields for Single View Reconstruction (CVPR 2023)

https://fwmb.github.io/bts/

BSD 2-Clause "Simplified" License

250 stars 19 forks source link

Does larger MLP affects the final results? #30

Closed ruili3 closed 1 year ago

ruili3 commented 1 year ago

Hi Brummi,

Thanks a lot for the awesome code! I noticed you use a quite small MLP to render the density field, e.g., ResnetFC with small hidden dimension channels (64) and without any ResNet blocks (0).

I wonder if a larger MLP with 1) larger dimensions; 2) more layers will lead to suboptimal results according to your previous experiments.

Thanks a lot for the information :)

Brummi commented 1 year ago

Hi, sorry for the late reply!

Yes, I have found that having a small MLP is essential to having stable trainings. The encoder-decoder receives the training signal only through the MLP. With a larger MLP, the signal becomes more noisy. I believe that this is one of the key reasons why PixelNeRF does not generalize very well.