A novel single-view point cloud generation model for multimodal spatial representation

doi:10.1016/j.patcog.2026.114649

A novel single-view point cloud generation model for multimodal spatial representation

2026-09-06

0

PRE

AI

DOI：10.1016/j.patcog.2026.114649

Original

Original request for help

Save

Abstract

En 中文

• Proposes a 3D spatial perception solution for open urban embodied AI systems. • Solves performance degradation from camera drift and pose variation in real apps. • Designs a spatial-aware architecture with human-like attention and multimodal fusion. • Achieves 40% point cloud quality gain on KITTI with +11.9M params, +30 ms latency.

Keywords:

Embodied intelligence

Points generation

Single-view reconstruction

Multimodal transformer

Spatially sensitive convolution

A novel single-view point cloud generation model for multimodal spatial representation

Abstract

Journal

Organization