CAMEO: Correspondence-Attention Alignment for Multi-View Diffusion Models

2025-12-02

AI Summary

Multi-view diffusion models suffer from degraded geometric correspondence and insufficient view consistency under large viewpoint variations in novel view synthesis. We reveal that their attention maps implicitly learn cross-view geometric correspondences during training, yet this signal remains sparse and unstable. To address this, we propose CAMEO: a lightweight, model-agnostic method that imposes explicit geometric supervision on only a single layer of self-attention maps—requiring no auxiliary networks or human annotations. By enhancing structural alignment across views, CAMEO reduces required training iterations by 50% on multiple benchmarks while improving PSNR by 1.2–2.3 dB at equal iteration counts. The generated views exhibit superior geometric consistency and visual quality, achieving state-of-the-art performance.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Multi-view diffusion models have recently emerged as a powerful paradigm for novel view synthesis, yet the underlying mechanism that enables their view-consistency remains unclear. In this work, we first verify that the attention maps of these models acquire geometric correspondence throughout training, attending to the geometrically corresponding regions across reference and target views for view-consistent generation. However, this correspondence signal remains incomplete, with its accuracy degrading under large viewpoint changes. Building on these findings, we introduce CAMEO, a simple yet effective training technique that directly supervises attention maps using geometric correspondence to enhance both the training efficiency and generation quality of multi-view diffusion models. Notably, supervising a single attention layer is sufficient to guide the model toward learning precise correspondences, thereby preserving the geometry and structure of reference images, accelerating convergence, and improving novel view synthesis performance. CAMEO reduces the number of training iterations required for convergence by half while achieving superior performance at the same iteration counts. We further demonstrate that CAMEO is model-agnostic and can be applied to any multi-view diffusion model.

Problem

• Enhances multi-view diffusion models' geometric correspondence accuracy
• Reduces training iterations by half while improving synthesis quality
• Provides a model-agnostic training technique for view-consistent generation

Innovation

• Supervises attention maps using geometric correspondence
• Guides model to learn precise correspondences in single layer
• Reduces training iterations by half while improving performance