Panorama Image Completion via Diffusion

Paper Title:

IPO-LDM: Depth-aided 360-degree Indoor RGB Panorama Outpainting via Latent Diffusion Model

Published on:

6 July 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Tianhao Wu,

Chuanxia Zheng,

Tat-Jen Cham

Bullets

Key Details

•

Proposes panorama completion via latent diffusion models, conditioned on visible pixels

•

Uses depth during training to improve spatial reasoning, not needed at test time

•

Achieves wraparound consistency via camera rotation and alignment

•

Significantly outperforms state-of-the-art methods on diverse metrics

•

Produces varied, realistic indoor layouts and objects

Explore the topics in this paper

diffusion models

latent representations

panorama generation

rgb images

scene understanding

AI generated summary

Panorama Image Completion via Diffusion

This paper proposes a new method to generate complete 360-degree RGB panorama images from partial narrow field-of-view inputs, using latent diffusion models. It introduces depth information during training to aid spatial understanding, and novel techniques to achieve wraparound consistency. Results significantly outperform prior work in generating diverse, realistic indoor scenes.