NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis
AuthorsRamil Khafizov, Ilya Statsenko, Ruslan Rakhimov, Artem Komarichev, Peter Wonka, Evgeny Burnaev
AffiliationsApplied AI Institute · T-Tech · KAUST · AXXX
NAMVIS makes multi-view image generation faster by using camera-aware autoregressive scaling instead of slow diffusion denoising.
Key results
NAMVIS transformer parameter count.
Approximate number of filtered Objaverse-XL objects.
NAMVIS result on the in-domain benchmark.
Times faster than the fastest evaluated diffusion baseline in the single-view-to-single-view setting.
Average reconstructed points from generated multi-view outputs.
What the paper found
NAMVIS tackles sparse-view novel-view synthesis without the repeated denoising used by diffusion-based approaches built on models such as Stable Diffusion. Built on Infinity, its 1B-parameter transformer generates discrete visual tokens from coarse to fine, predicting tokens in parallel within each scale and across target views. Its key technique, Multi-scale Projective Pose Encoding, injects camera geometry into both target-view self-attention and source-to-target cross-attention; a dual conditioning path combines global source-image context with dense spatial features from a frozen VQ-VAE. Trained on approximately 200K filtered Objaverse-XL objects, NAMVIS is evaluated on Objaverse, GSO, and OmniObject3D at 256 × 256 resolution. On Objaverse it achieves 22.485 PSNR, and in the single-source, single-target setting it generates a view in 0.6 seconds—about 3.3× faster than the fastest evaluated diffusion baseline. It also produces views that are more amenable to 3D reconstruction, yielding 221 average COLMAP sparse points. The results position geometry-conditioned next-scale autoregression as a faster alternative to diffusion for multi-view image synthesis.
Original abstract
Sparse-view novel view synthesis is a central problem in 3D content creation, but diffusion-based approaches remain limited by iterative denoising, making multi-view generation expensive at inference time. We introduce NAMVIS, a diffusion-free framework that reformulates multi-view image synthesis as geometry-conditioned next-scale autoregression. Instead of generating target views through repeated denoising, NAMVIS predicts discrete visual tokens through a small number of coarse-to-fine scale steps, while sampling all tokens within each scale and across target views in parallel. To anchor this generation process to explicit camera geometry, we propose Multi-scale Projective Pose Encoding, which injects source and target camera transformations into both target-view self-attention and source-to-target cross-attention at every resolution. NAMVIS further combines global conditioning with dense geometry-aware cross-attention, enabling the model to preserve source-view appearance while maintaining target-view consistency. Across Objaverse, GSO, and OmniObject3D, NAMVIS outperforms diffusion-based baselines in PSNR, SSIM, and LPIPS, while running over 3 times faster than the evaluated diffusion baselines under the same evaluation setting. These results suggest that geometry-conditioned next-scale autoregression is a promising and efficient alternative to diffusion for sparse-view multi-view synthesis. Additional qualitative results, videos, and resources are available at https://corl-team.github.io/namvis/
Read the original paperMore in Computer Vision
Browse all 64 papers →LEGO: A Lifting-Free Approach for Exocentric-to-Egocentric Video Generation
Suhwan Cho, Yonwoo Choi, Soongjin Kim, Jicheol Park, Taegyu Lim
LEGO replaces brittle depth-based camera lifting with a learned transformer view synthesizer that gives diffusion models better structure for generating first-person video.
VibeEdit: Image Editing with Canvas Instructions
Jinjing Zhao, Fangyun Wei, Yitong Wang, Xiuyu Wu, Yunuo Chen, Yang Yue, Sirui Zhang, Wenbo Wang, Hongyang Zhang, Dong Chen, Yan Lu, Chang Xu
VibeEdit lets users edit the exact objects they mean by marking them directly on an image and briefly describing the desired change.
JLD: Perceptual Distance Through A Jacobian Lens
Shreshth Saini, Balu Adsumilli, Alan C. Bovik
JLD uses the Jacobian of a frozen vision model to build a faster, more resolution-robust measure of how visually different two images appear.