Geometry Fidelity for Spherical Images
Abstract
Spherical or omni-directional images offer an immersive visual format appealing to a wide range of computer vision applications. However, geometric properties of spherical images pose a major challenge for models and metrics designed for ordinary 2D images. We show that direct application of Fréchet Inception Distance (FID) is insufficient for quantifying geometric fidelity in spherical images. To remedy this, we introduce Omnidirectional FID (OmniFID), an extension of FID, which additionally captures field-of-view requirements of the spherical format.
Cite
Text
Christensen et al. "Geometry Fidelity for Spherical Images." ICML 2024 Workshops: GRaM, 2024.Markdown
[Christensen et al. "Geometry Fidelity for Spherical Images." ICML 2024 Workshops: GRaM, 2024.](https://mlanthology.org/icmlw/2024/christensen2024icmlw-geometry/)BibTeX
@inproceedings{christensen2024icmlw-geometry,
title = {{Geometry Fidelity for Spherical Images}},
author = {Christensen, Anders and Mojab, Nooshin and Patel, Khushman and Ahuja, Karan and Akata, Zeynep and Winther, Ole and Gonzalez-Franco, Mar and Colaco, Andrea},
booktitle = {ICML 2024 Workshops: GRaM},
year = {2024},
url = {https://mlanthology.org/icmlw/2024/christensen2024icmlw-geometry/}
}