Hash Grid Feature Pruning

TL;DR

This paper introduces a hash grid feature pruning method for Gaussian splatting that reduces storage and transmission costs by removing invalid features, improving rate-distortion performance.

Contribution

It presents a novel pruning technique that identifies and removes invalid hash grid features based on Gaussian splat coordinates, optimizing storage efficiency.

Findings

Achieves an average bitrate reduction of 8% under standard conditions.

Reduces hash grid storage size without affecting model performance.

Improves rate-distortion efficiency in Gaussian splatting applications.

Abstract

Hash grids are widely used to learn an implicit neural field for Gaussian splatting, serving either as part of the entropy model or for inter-frame prediction. However, due to the irregular and non-uniform distribution of Gaussian splats in 3D space, numerous sparse regions exist, rendering many features in the hash grid invalid. This leads to redundant storage and transmission overhead. In this work, we propose a hash grid feature pruning method that identifies and prunes invalid features based on the coordinates of the input Gaussian splats, so that only the valid features are encoded. This approach reduces the storage size of the hash grid without compromising model performance, leading to improved rate-distortion performance. Following the Common Test Conditions (CTC) defined by the standardization committee, our method achieves an average bitrate reduction of 8% compared to the baseline approach.