Ref-GeNVS: Training-Free Reflection-Aware Generative Novel View Synthesis for Mirror Scenes

Forum topic · 小凯 · 2026-09-09

Summary

Ref-GeNVS is a training-free, reflection-aware method for generative novel view synthesis (NVS) in scenes containing mirrors, proposed by GeonU Kim, Shin Dong-Yeon, and Tae-Hyun Oh (arXiv:2609.05382). Existing multi-view diffusion models often fail to recognize mirrors and cannot exploit reflected content when generating new views. Ref-GeNVS addresses this without additional training by treating a mirror image as two complementary views: it estimates the mirror plane from input images and reflects camera poses to form virtual views. On this setup, a two-stage generation process—Mirror-gated attention and Reflection injection—explicitly leverages reflection relationships inside a multi-view diffusion model to achieve reflection-consistent synthesis. The method inherits the strong generalizability of the multi-view diffusion backbone with no fine-tuning, and outperforms recent generative NVS approaches on synthetic and real mirror scenes, revealing scene structures visible only through reflections.

Research area: Computer Vision Authors: GeonU Kim, Shin Dong-Yeon, Tae-Hyun Oh Published: 2026-09-04 arXiv: 2609.05382

Overview

This paper proposes Ref-GeNVS, a training-free, reflection-aware method for generative novel view synthesis (NVS) in mirror scenes.

Problem

Existing multi-view diffusion models often fail to recognize the mirror in the scene and cannot exploit reflected content for scene generation.

Key Idea

To fix this issue without additional training, the authors treat a mirror image as two complementary views:

From input images, estimate the mirror plane and reflect camera poses to form virtual views.
Based on this virtual view setup, apply a two-stage generation method:

Mirror-gated attention

Reflection injection

These stages enable reflection-consistent NVS by explicitly leveraging reflection relationships within a multi-view diffusion model.

Results

Ref-GeNVS inherits the strong generalizability of the multi-view diffusion backbone — no fine-tuning required.
On both synthetic and real scenes containing mirrors, it outperforms recent generative NVS methods by producing reflection-consistent and contextually coherent novel views.
It can reveal scene structures that are visible only through the mirror.

Original Abstract (excerpt)

> We propose Ref-GeNVS, a training-free, reflection-aware method for generative novel view synthesis (NVS) in mirror scenes. Existing multi-view diffusion models often fail to recognize the mirror in the scene and cannot exploit reflected content for scene generation. To fix this issue without additional training, our key idea is to treat a mirror image as two complementary views... Ref-GeNVS inherits the strong generalizability of the multi-v...

---

*Auto-collected on 2026-09-09.*