Reflection-aware Generative Novel View Synthesis

GeonU Kim; Shin Dong-Yeon; Tae-Hyun Oh

cs.CVSep 4, 2026

Reflection-aware Generative Novel View Synthesis

Authors: GeonU Kim, Shin Dong-Yeon, Tae-Hyun Oh

Organizations: KAIST

Abstract

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. From input images, we estimate the mirror plane and reflect camera poses to form virtual views. Based on this virtual view setup, we propose a two-stage generation method consisting of Mirror-gated attention and Reflection injection, which enables reflection-consistent NVS by explicitly leveraging reflection relationships in a multi-view diffusion model. Ref-GeNVS inherits the strong generalizability of the multi-view diffusion backbone, while it does not require finetuning. On synthetic and real scenes including mirrors, Ref-GeNVS outperforms recent generative NVS methods by generating reflection-consistent and contextually coherent novel views, revealing scene structure visible only through mirrors. Project page: https://kim-geonu.github.io/Ref-GeNVS/

Figures & tables

Figure 2: Overview of Ref-GeNVS. In Stage 1, we first mask mirror surface to prevent interference between real-scene and reflected geometry. Then we generate the reflected scene by mirror-masked inputs and the reflected virtual views which consist of flipped input images, reflected camera poses, and the mirror masks. We propose Mirror-gated attention enabling only pixels inside the mirror influence the reflected views. In Stage 2, we perform Reflection injection process which generates the mirror surface using the images generated in Stage 1 as inputs. Finally, we obtain complete target images $\textbf{I}_{o}$ from target views, which are consistent with the scene reflected on the mirror.

Figure 3: Reflection injection process in Stage 2. We fill the mirror surface in the target views using three sources: the original inputs $(\textbf{I}_{s},\textbf{P}_{s})$ , the reflected inputs from the preprocess $(\textbf{I}^{\prime\,m}_{s},\textbf{P}^{\prime\,m}_{s},\textbf{M}^{\prime\,m}_{s})$ , and the Stage 1 outputs $(\textbf{I}_{o}^{\text{mask}},\textbf{P}_{o})$ . To generate the mirror surface in the Stage 1 outputs $(\textbf{I}_{o}^{\mathrm{mask}\_m},\textbf{P}_{o}^{m})$ consistent with the scene in reflection, we condition the model on two type of target-frame poses: the identity poses $\textbf{P}_{o}^{m}$ and the corresponding reflected poses $\textbf{P}^{\prime\,m}_{o}$ . During each denoising step $t\!\to\!t\!-\!1$ , the latent feature maps $\textbf{Z}^{\prime\,m}_{t}$ predicted under $\textbf{P}^{\prime\,m}_{o}$ are horizontally flipped and injected inside the target mirror mask $\tilde{\textbf{M}}_{o}^{m}$ of the current target latent $\textbf{Z}^{\,m}_{t}$ . This latent injection is integrated with SDEdit-style guidance [ 37 ] , enabling natural mirror-surface completion. The result is the inpainted mirror which is consistent with the scene generated in Stage 1 and with the reflected evidence from the original input images.

Figure 4: Effect of SDEdit-style [ 37 ] guidance. Without SDEdit-style guidance, the model can produce noticeable artifacts near the boundaries of the mirror. The guidance enable natural compositing of mirror and non-mirror region.

Figure 5: Automatic preprocess. We first segment mirror from the input images using DAM [ 62 ] and mask mirror with a uniform color. Then we estimate the mirror plane from input images by first regressing per-pixel 3D points [ 32 ] and obtaining the plane equation of the mirror using RANSAC [ 18 ] algorithm. We then reflect the input camera poses across the plane to obtain the reflected camera poses.

Figure 6: Input views and test trajectories. To evaluate reflection-aware novel view synthesis, we sample challenging, scene-specific trajectories for our synthetic dataset (top) and the real dataset [ 67 ] (bottom). Each input view contains a mirror, and the trajectory includes regions visible only via reflection (red dashed box). For clarity, we overlay neon highlights to mark the mirror in the input views.

	Real data [ 67 ]					Synthetic data
Metric	DS $\downarrow$	CLIP $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	DS $\downarrow$	CLIP $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
MVGenMaster [ 6 ]	0.361	0.867	12.75	0.413	0.536	0.227	0.926	16.32	0.719	0.378
Ours (MVGen.)	0.194	0.930	13.67	0.455	0.474	0.116	0.959	17.50	0.732	0.313
SEVA [ 70 ]	0.275	0.920	12.03	0.394	0.564	0.270	0.924	13.62	0.612	0.492
Ours (SEVA)	0.129	0.951	14.19	0.455	0.466	0.156	0.948	15.27	0.676	0.403

Table 1: Quantitative results of sparse-image novel view synthesis using DreamSim (DS) [ 20 ] , CLIP [ 41 ] similarity score, PSNR, SSIM [ 58 ] , and LPIPS [ 69 ] . Our method achieves better performance compared to MVGenMaster [ 6 ] and SEVA [ 70 ] .

Figure 7: Qualitative comparison under sparse-input novel view synthesis. Ref-GeNVS produces more consistent and reflection-aware target views compared to prior methods [ 70 , 6 ] , especially in regions only visible through the mirror.

Figure 8: Qualitative comparison in the single image NVS setting. Despite observing only single input image, Ref-GeNVS utilizes mirror cues to generate coherent novel views, outperforming prior methods [ 54 , 10 , 70 ] on reflected and hidden regions.

Method	DS $\downarrow$	CLIP $\uparrow$
VistaDream [ 54 ]	0.569	0.866
FlexWorld [ 10 ]	0.369	0.879
SEVA [ 70 ]	0.377	0.905
Ours	0.341	0.911

Table 2: Single image result.

Virtual view	Mirror-gated attn.	Two-step gen.	DS $\downarrow$	CLIP $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
X	X	X	0.270	0.924	13.62	0.612	0.492
O	X	X	0.229	0.923	13.59	0.643	0.536
O	O	X	0.160	0.948	15.01	0.671	0.416
O	O	O	0.156	0.948	15.27	0.676	0.403

Table 3: Ablation study on injection of reflected virtual view, Mirror-gated attention (Mirror-gated attn.), and two-step generation process (Two-step gen.). We compare the performance by adding each component on the baseline model, SEVA [ 70 ] .

Figure 9: Ablation on synthetic scenes by progressively adding our components: (a) SEVA baseline, (b) + virtual view, (c) + Mirror-gated attention, and (d) + two-stage generation (full method). Our full model produces reflections consistent with the generated scene.

Figure 10: Qualitative comparison of 3D reconstruction with generated images using $\text{Pi}^{3}$ [ 57 ] . Ref-GeNVS produces more complete and stable geometry in mirror-containing scenes compared to FlexWorld [ 10 ] and SEVA [ 70 ] , with a single input image.

Figure 11: Result using an input with a distorted mirror. We warp the distorted reflection into the planar-mirror representation and apply Ref-GeNVS without modification.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Figure S1 : 3D point prediction with Dust3R [ 56 ] , $\text{Pi}^{3}$ [ 57 ] , and Depth Anything 3 (DA3) [ 32 ] on original images (left) and mirror-masked images (right). $\text{Pi}^{3}$ and DA3 randomly predict 3D points of mirror surface according to the input image: as flat plane or as a hole in the scene geometry. After masking the mirror, they consistently predict the mirror as a flat plane.

Figure S2 : Robustness analysis under imperfect mirror masks. We corrupt the mirror region by pasting random patches before DAM-based mirror segmentation and measure the IoU between the predicted mirror mask and the ground-truth mirror mask. The x-axis shows the mask IoU and the y-axis shows reconstruction performance measured by DreamSim, LPIPS, CLIP similarity, and PSNR. We compare two mirror plane estimation strategies: an edge-based method (Edge) that fits the mirror plane from edge points without mirror masking, and our surface-based method (Surface) that applies mirror masking and estimates the plane from surface points predicted by per-pixel 3D regression. As mask IoU decreases, the performance of both methods degrades, but the proposed surface-based estimation remains consistently more robust than the edge-based approach. The dashed line indicates the baseline NVS model (SEVA).

Figure S3 : Failure case of edge-based mirror plane estimation under imperfect mirror masks. Given input images, mirror masks estimated by DAM contain inaccuracies. When mirror plane estimation relies on edge points, the inaccurate mask leads to incorrect plane estimation and misaligned reflected content. In contrast, our surface-based mirror plane estimation remains robust by fitting the plane using surface points predicted from per-pixel 3D regression, producing reflection-consistent novel views.

Figure S4 : Failure case. In extreme cases where point cloud regression is unreliable ( e.g . the mirror is only partially visible), inaccurate mirror-plane estimates can produce results that are misaligned with the ground truth at the same camera poses.

Figure S5 : Additional qualitative comparison of sparse images novel view synthesis with the prior methods [ 70 , 6 ] . The left-most scene is from the real dataset [ 67 ] and the middle and right scenes are from our synthetic dataset.

Figure S6 : Additional qualitative results of single image novel view synthesis on the real-world dataset, the mirror segmentation dataset (MSD) [ 63 ] (left and middle) and the MMD dataset [ 59 ] (right). There is no ground truth scene images since the datasets are originally designed for mirror segmentation evaluation.

Method	Full	Reflected region	Mirror surface
MVGenMaster	16.90	16.51	19.41
Ours (MVGen.)	17.95 (+1.05)	18.40 (+1.89)	19.69 (+0.28)
SEVA	13.65	13.31	15.25
Ours (SEVA)	15.51 (+1.86)	16.11 (+2.80)	16.16 (+0.91)

Appendix

Table S1 : PSNR comparison on the six scenes of synthetic data with sparse-input setting. Our method shows larger performance gap in the reflected region than in the full scene.

Figure S7 : Examples of reflection mask and mirror mask. We extract ground truth binary masks of reflected region of the scene and mirror region using Blender [ 12 ] .

Explore similar work

BetterScene: 3D Scene Synthesis with Representation-Aligned Generative Model
Feb 26, 2026Yuci Han, Charles Toth, John E. Anderson +2Novel View Synthesis Sparse-View 3D Reconstruction
Dept. of Electrical and Computer Engineering, The Ohio State University · USACE ERDC GRL
MirrorWorld: Taming Video Diffusion Models for Mirror Reflection Generation
Aug 7, 2026Youjun Zhao, Alex Warren, Gary K. L. Tam +1Video Diffusion Models Video Generation
City University of Hong Kong · Swansea University
GenRec: Knowing Where to Reconstruct and Where to Generate
Aug 18, 2026Ata Çelen, Jaewoo Jung, Federico Tombari +4Novel View Synthesis Sparse-View 3D Reconstruction
ETH Zürich · KAIST · Google +1

Feb 26, 2026J/K move · Enter open · S save

BetterScene: 3D Scene Synthesis with Representation-Aligned Generative Model

Yuci Han, Charles Toth, John E. Anderson, William J. Shuart+1

Dept. of Electrical and Computer Engineering, The Ohio State University · USACE ERDC GRL

Novel View Synthesis Sparse-View 3D Reconstruction Video Diffusion Models 3D Reconstruction

Open paper AskPDF arXiv

We present BetterScene, an approach to enhance novel view synthesis (NVS) quality for diverse real-world scenes using extremely sparse, unconstrained photos. BetterScene leverages the production-ready Stable Video Diffusion (SVD) model pretrained on billions of frames as a strong backbone, aiming to mitigate artifacts and recover view-consistent details at inference time. Conventional methods have developed similar diffusion-based solutions to address these challenges of novel view synthesis. Despite significant improvements, these methods typically rely on off-the-shelf pretrained diffusion priors and fine-tune only the UNet module while keeping other components frozen, which still leads to inconsistent details and artifacts even when incorporating geometry-aware regularizations like depth or semantic conditions. To address this, we investigate the latent space of the diffusion model and introduce two components: (1) temporal equivariance regularization and (2) vision foundation model-aligned representation, both applied to the variational autoencoder (VAE) module within the SVD pipeline. BetterScene integrates a feed-forward 3D Gaussian Splatting (3DGS) model to render features as inputs for the SVD enhancer and generate continuous, artifact-free, consistent novel views. We evaluate on the challenging DL3DV-10K dataset and demonstrate superior performance compared to state-of-the-art methods.

Abstract

Explore similar work

BetterScene: 3D Scene Synthesis with Representation-Aligned Generative Model

MirrorWorld: Taming Video Diffusion Models for Mirror Reflection Generation

GenRec: Knowing Where to Reconstruct and Where to Generate