3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Oh, Gyeongrok, Kim, Sungjune, Ko, Heeju, Chi, Hyung-gun, Kim, Jinkyu, Lee, Dongwook, Ji, Daehyun, Choi, Sungjoon, Jang, Sujin, Kim, Sangpil
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916656872161280
author Oh, Gyeongrok
Kim, Sungjune
Ko, Heeju
Chi, Hyung-gun
Kim, Jinkyu
Lee, Dongwook
Ji, Daehyun
Choi, Sungjoon
Jang, Sujin
Kim, Sangpil
author_facet Oh, Gyeongrok
Kim, Sungjune
Ko, Heeju
Chi, Hyung-gun
Kim, Jinkyu
Lee, Dongwook
Ji, Daehyun
Choi, Sungjoon
Jang, Sujin
Kim, Sangpil
contents The resolution of voxel queries significantly influences the quality of view transformation in camera-based 3D occupancy prediction. However, computational constraints and the practical necessity for real-time deployment require smaller query resolutions, which inevitably leads to an information loss. Therefore, it is essential to encode and preserve rich visual details within limited query sizes while ensuring a comprehensive representation of 3D occupancy. To this end, we introduce ProtoOcc, a novel occupancy network that leverages prototypes of clustered image segments in view transformation to enhance low-resolution context. In particular, the mapping of 2D prototypes onto 3D voxel queries encodes high-level visual geometries and complements the loss of spatial information from reduced query resolutions. Additionally, we design a multi-perspective decoding strategy to efficiently disentangle the densely compressed visual cues into a high-dimensional 3D occupancy scene. Experimental results on both Occ3D and SemanticKITTI benchmarks demonstrate the effectiveness of the proposed method, showing clear improvements over the baselines. More importantly, ProtoOcc achieves competitive performance against the baselines even with 75\% reduced voxel resolution.
format Preprint
id arxiv_https___arxiv_org_abs_2503_15185
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle 3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation
Oh, Gyeongrok
Kim, Sungjune
Ko, Heeju
Chi, Hyung-gun
Kim, Jinkyu
Lee, Dongwook
Ji, Daehyun
Choi, Sungjoon
Jang, Sujin
Kim, Sangpil
Computer Vision and Pattern Recognition
Artificial Intelligence
The resolution of voxel queries significantly influences the quality of view transformation in camera-based 3D occupancy prediction. However, computational constraints and the practical necessity for real-time deployment require smaller query resolutions, which inevitably leads to an information loss. Therefore, it is essential to encode and preserve rich visual details within limited query sizes while ensuring a comprehensive representation of 3D occupancy. To this end, we introduce ProtoOcc, a novel occupancy network that leverages prototypes of clustered image segments in view transformation to enhance low-resolution context. In particular, the mapping of 2D prototypes onto 3D voxel queries encodes high-level visual geometries and complements the loss of spatial information from reduced query resolutions. Additionally, we design a multi-perspective decoding strategy to efficiently disentangle the densely compressed visual cues into a high-dimensional 3D occupancy scene. Experimental results on both Occ3D and SemanticKITTI benchmarks demonstrate the effectiveness of the proposed method, showing clear improvements over the baselines. More importantly, ProtoOcc achieves competitive performance against the baselines even with 75\% reduced voxel resolution.
title 3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2503.15185