MultiViewPano: A Generalist Approach to 360-degree Panorama Generation

Simon Coessens, Akash Malhotra, Nacera seghouani

Keywords: 360° Panorama Generation, Multi-View Diffusion, Novel View Synthesis, Image Stitching, Pose-Aware Blending, Training-Free Methods, View Consistency

TL;DR: We propose MultiViewPano, a training-free pipeline that uses a pretrained multi-view diffusion model and a pose-aware stitcher to generate high-quality 360° panoramas from one or more arbitrarily posed input images.

Abstract: We propose MultiViewPano, a training-free framework for 360° panorama generation from one or more arbitrarily posed input images. Our approach leverages a pretrained multi-view diffusion model to synthesize novel views along a virtual camera trajectory, which are then fused using a custom pose-aware stitching pipeline. Unlike prior methods that require fixed field-of-view inputs or task-specific fine-tuning, MultiViewPano supports flexible camera poses and generalizes across diverse scenes. Our experiments demonstrate that our method achieves competitive visual fidelity compared to state-of-the-art approaches, while offering greater flexibility and simplicity.

Submission Number: 22