Beyond Global Latents: Chunk-Based Sparse Grid VAE for Scalable 3D Modeling
2026-08-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors present ChunkVAE, a new way to compress 3D data using small local pieces called chunks instead of one big global block. This makes it easier to work with very detailed 3D shapes without using too much memory. They use special methods to split and combine these chunks so the final 3D shape looks accurate and smooth. Their approach works well on several tests and runs faster because it can handle many chunks at the same time. Overall, their method helps keep detailed 3D structures manageable and useful for further tasks.
Sparse voxel gridsVariational autoencoder (VAE)3D reconstructionLocal latent spaceBinary object partitioningS-Curve stitchingMemory efficiencyParallel inferenceGeometry compression
Authors
Kaiyi Zhang, Zhihao Liang, Haolin Liu, Qingxiang Lin, Zeqiang Lai, Yunfei Zhao, Bowen Zhang, Xianghui Yang, Zibo Zhao, Chunchao Guo, Long Quan
Abstract
Sparse voxel grids preserve the spatial structure needed for detailed 3D reconstruction, but their memory still grows rapidly with resolution as active surface cells increase. We introduce ChunkVAE, a sparse grid variational autoencoder organized around local chunks rather than a global latent volume. Local learned operators permit independently chosen encoder and decoder partitions and allow inference chunk sizes to differ from training. Two complementary data operators make this flexibility practical: Balanced Binary Object Partitioning distributes active cells while limiting replicated overlap, while S-Curve weighted stitching attenuates unreliable boundary features when assembling a global latent or reconstruction. Across three object benchmarks, ChunkVAE is competitive with or better than strong baselines from $512^3$ to $1536^3$; smaller chunks lower peak allocated memory and shorten per-chunk compute, enabling faster parallel inference. Stable stitched latents and improved image to 3D metrics indicate that local compression can scale geometry while retaining the global interface required downstream.