HashEncoding: Autoencoding with Multiscale Coordinate Hashing

TL;DR

HashEncoding introduces a multiscale coordinate hashing autoencoder that enables efficient, parameter-light image reconstruction and geometric task performance by leveraging space-folding properties of hash functions.

Contribution

It proposes a novel autoencoding architecture using multiscale coordinate hashing, reducing decoder parameters and enabling direct backpropagation for geometric tasks.

Findings

Achieves near non-parametric image reconstruction with fewer parameters.

Enables effective optical flow estimation through backpropagation to coordinate space.

Demonstrates improved generalizability over traditional autoencoders.

Abstract

We present HashEncoding, a novel autoencoding architecture that leverages a non-parametric multiscale coordinate hash function to facilitate a per-pixel decoder without convolutions. By leveraging the space-folding behaviour of hashing functions, HashEncoding allows for an inherently multiscale embedding space that remains much smaller than the original image. As a result, the decoder requires very few parameters compared with decoders in traditional autoencoders, approaching a non-parametric reconstruction of the original image and allowing for greater generalizability. Finally, by allowing backpropagation directly to the coordinate space, we show that HashEncoding can be exploited for geometric tasks such as optical flow.