https://github.com/tancik/learnit

https://github.com/tancik/learnit

Last synced: 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/tancik/learnit
• Owner: tancik
• License: mit
• Created: 2021-01-01T21:56:56.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-07-03T16:52:34.000Z (over 3 years ago)
• Last Synced: 2024-08-01T13:23:21.520Z (5 months ago)
• Language: Jupyter Notebook
• Size: 4.84 MB
• Stars: 157
• Watchers: 9
• Forks: 27
• Open Issues: 5
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Learned Initializations for Optimizing Coordinate-Based Neural Representations

### [Project Page](https://www.matthewtancik.com/learnit) | [Paper](https://arxiv.org/abs/2012.02189)

[![Open Demo in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/tancik/learnit/blob/master/meta_demo.ipynb)


[Matthew Tancik](http://tancik.com/)\*1,

[Ben Mildenhall](https://people.eecs.berkeley.edu/~bmild/)\*1,

Terrance Wang1,

Divi Schmidt1,

[Pratul P. Srinivasan](https://people.eecs.berkeley.edu/~pratul/)2,

[Jonathan T. Barron](http://jonbarron.info/)2,

[Ren Ng](https://www2.eecs.berkeley.edu/Faculty/Homepages/yirenng.html)1


1UC Berkeley, 2Google Research

*denotes equal contribution

## Abstract

![Teaser Image](https://user-images.githubusercontent.com/3310961/103447030-36275800-4c43-11eb-92ac-a242130d6e04.jpg)

Coordinate-based neural representations have shown significant promise as an alternative to discrete, array-based representations for complex low dimensional signals. However, optimizing a coordinate-based network from randomly initialized weights for each new signal is inefficient. We propose applying standard meta-learning algorithms to learn the initial weight parameters for these fully-connected networks based on the underlying class of signals being represented (e.g., images of faces or 3D models of chairs). Despite requiring only a minor change in implementation, using these learned initial weights enables faster convergence during optimization and can serve as a strong prior over the signal class being modeled, resulting in better generalization when only partial observations of a given signal are available.

## Code

We provide a [demo IPython notebook](https://colab.research.google.com/github/tancik/learnit/blob/master/meta_demo.ipynb) as a simple reference for the core idea. Scripts for the different tasks are located in the [Experiments](https://github.com/tancik/learnit/tree/master/Experiments) directory.