Reflection-aware Generative Novel View Synthesis
Ref-GeNVS treats mirror reflections as usable scene views, without retraining the diffusion model.
The method estimates the mirror plane from input images, then reflects camera poses to create virtual views. It uses mirror-gated attention and reflection injection to keep generated novel views consistent with reflected content. The authors say it outperforms recent generative NVS methods on synthetic and real mirror scenes, including cases where structure is visible only through the mirror. ArXiv · AI/CL/LG's note
The method estimates the mirror plane from input images, then reflects camera poses to create virtual views. It uses mirror-gated attention and reflection injection to keep generated novel views consistent with reflected content. The authors say it outperforms recent generative NVS methods on synthetic and real mirror scenes, including cases where structure is visible only through the mirror. ArXiv · AI/CL/LG's note
score 4