RocNet: Recursive Octree Network for Efficient 3D Deep Representation

TL;DR

RocNet is a recursive octree-based deep neural network that efficiently compresses 3D voxel data into a small latent space, enabling accurate shape classification, reconstruction, and generation with reduced memory and training time.

Contribution

The paper introduces RocNet, a novel recursive octree network that significantly improves 3D data compression and processing efficiency over existing methods.

Findings

Compresses 3D voxel grids to 80 floats in latent space

Maintains accuracy in shape classification, reconstruction, and generation

Reduces memory usage and training time compared to prior approaches

Abstract

We introduce a deep recursive octree network for the compression of 3D voxel data. Our network compresses a voxel grid of any size down to a very small latent space in an autoencoder-like network. We show results for compressing 32, 64 and 128 grids down to just 80 floats in the latent space. We demonstrate the effectiveness and efficiency of our proposed method on several publicly available datasets with three experiments: 3D shape classification, 3D shape reconstruction, and shape generation. Experimental results show that our algorithm maintains accuracy while consuming less memory with shorter training times compared to existing methods, especially in 3D reconstruction tasks.