Multigranularity-3DQA: Dynamic multi-granularity perception and gated dual-stage encoder-decoder for 3D question answering

Letao Zhang

doi:10.1016/j.eswa.2026.131676

Multigranularity-3DQA: Dynamic multi-granularity perception and gated dual-stage encoder-decoder for 3D question answering

2026-02-24

0

PRE

AI

DOI：10.1016/j.eswa.2026.131676

Original

Original request for help

Save

Abstract

En 中文

• Dynamic Multi-Granularity Fusion enhances fine-grained feature extraction in 3D point clouds • Adaptive Region-Aware Attention captures spatial relationships through variable attention fields • Gated dual-stage encoder-decoder architecture enables precise cross-modal feature fusion • State-of-the-art performance on ScanQA dataset with enhanced small object perception

Keywords:

Dynamic Multi-Granularity Fusion

Adaptive Region-Aware Attention

Gated Dual-Stage Encoder-Decoder

3D Question Answering

Fine-Grained Feature Extraction

Multigranularity-3DQA: Dynamic multi-granularity perception and gated dual-stage encoder-decoder for 3D question answering

Abstract

Journal

Organization