robot0321 / learnit_exp

some experiments for "learned initializations for optimizing coordinate-based neural representations"

Learned Initializations for Optimizing Coordinate-Based Neural Representations

Forked from https://github.com/tancik/learnit

You should follow the instruction described in the above link.

Note) for cvlab@snu members, the data is placed in zeus server

This repo contains my own experiments on the meta-learning of implicit neural representations (especially for SIREN)

Experiment Description

• 2D image [celebA, IU, Anime, etc] -> Experiments/shapenet.ipynb
• 3D objet [shapenet, zju_mocap(human), etc] -> Experiments/shapenet.ipynb
• 3D NeRF (lego) -> Experiments/unisurf_meta.ipynb
• partial network meta-learning -> Experiments/freezeit...ipynb
• phototourism (from original work) -> Experiments/phototourism.ipynb

Note) Not all files are meaningful. (There are many dummy files.)
Further, among the files mentioned above, there are also meaningless blocks.
So please interpret them carefully...

2D image

3D object

• meta-initialization on 3D object