cs.CVNov 16, 2026

Efficient Multi-View Generation with Iterative Global-Local Proxy Attention

Authors: Wei Tan, Kunming Luo, Hongyu Yan, Yu Zhang, Zihao Zhang, Xiatian Zhu, Lei Xie

Organizations: School of Cyber Science and Engineering, Southeast University, Nanjing 210096, China. · Department of Electronic and Computer Engineering, the Hong Kong University of Science and Technology, Hong Kong, China. · College of Electronic and Information Engineering, Southwest University, Chongqing, China. · Institute of AI for Industries, Chinese Academy of Sciences, Nanjing, China. · 

Abstract

Generating consistent multi-view RGB images and normal maps from a single input image provides complementary appearance and geometric cues for 3D reconstruction. However, maintaining cross-view consistency requires information exchange across viewpoints, and dense cross-view attention becomes computationally expensive as the number of views and spatial resolution increase. In this paper, we propose Global-Local Proxy Attention (GLPA) to reduce cross-view communication costs while preserving fine-grained within-view processing. Specifically, GLPA retains full-resolution interactions within each view and uses compact global proxies for cross-view communication. To further leverage the exchanged information, GLPA incorporates iterative proxy refinement, which rebuilds the proxies from updated features and reuses the same attention parameters for subsequent interactions within a single denoising step. We evaluate GLPA on two structurally different diffusion backbones, FLUX and Stable Diffusion. Experiments on FLUX show that GLPA achieves a 2.65×2.65\times speedup in attention computation per denoising step over dense attention, while improving reconstruction IoU from 0.6849 to 0.7019 and reducing NVS LPIPS from 0.086 to 0.076. Results on Stable Diffusion further demonstrate the effectiveness of GLPA, with a 1.80×1.80\times attention speedup and improvements across all five quality metrics.

Figures & tables

Explore similar work

CardsList
  1. WorldRoPE: Probabilistic World-Centric Rotary Embedding for Multi-View Attention

    Oct 12, 2026Mijeong Kim, Kyujin Lim, Bohyung HanRotary Positional EmbeddingsMulti-View 3D Reconstruction

  2. GeoQuery: Geometry-Query Diffusion for Sparse-View Reconstruction

    May 12, 2026Xiao Cao, Yuze Li, Youmin Zhang +43D Gaussian SplattingDiffusion Models for Image Restoration

  3. TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention

    May 14, 2026David Huang, Guile Wu, Chengjie Huang +23D ViTsMulti-View 3D Reconstruction