thumbnail

PanoStyleVR: style-based similarity metrics for Web-based immersive panoramic style transfer

Muhammad Tukur, Sara Jashari, Fabio Bettio, Giovanni Pintore, Enrico Gobbetti, Jens Schneider, and Marco Agus

September 2025

Abstract

We introduce PanoStyleVR, an immersive web-based framework for analyzing, ranking, and interactively applying style similarities within panoramic indoor scenes, enabling stereoscopic virtual exploration and photorealistic style adaptation. A key innovation of our system is a fully immersive WebXR interface, allowing users wearing head-mounted displays to navigate indoor environments in stereo and apply new styles in real time. Style suggestions are visualized through floating thumbnails rendered in the VR space; selecting a style triggers photorealistic transfer on the current room view and updates the immersive stereo representation. This interactive pipeline is powered by two integrated neural components: (1) a geometry-aware and shading-independent GAN-based framework for semantic style transfer on albedo-reflectance representations; and (2) a gated architecture that synthesizes omnidirectional stereoscopic views from a single 360^CIRC panorama for realistic depth-aware exploration. Our system enables cosine-similarity-based style ranking, t-SNE-driven dimensionality reduction, and GMM-based clustering over large-scale panoramic datasets. These components support an immersive recommendation mechanism that connects stylistic analysis with interactive editing. Experimental evaluations on the Structured3D dataset demonstrate strong alignment between perceptual similarity and our proposed metric, and effective grouping of panoramas based on latent style representations.

Reference and download information

Muhammad Tukur, Sara Jashari, Fabio Bettio, Giovanni Pintore, Enrico Gobbetti, Jens Schneider, and Marco Agus. PanoStyleVR: style-based similarity metrics for Web-based immersive panoramic style transfer. In Proc. ACM Web3D, September 2025. To appear.

Related multimedia productions

Bibtex citation record

@inproceedings{Tukur:2025:PSS,
    author = {Muhammad Tukur and Sara Jashari and Fabio Bettio and Giovanni Pintore and Enrico Gobbetti and Jens Schneider and Marco Agus},
    title = {{PanoStyleVR}: style-based similarity metrics for Web-based immersive panoramic style transfer},
    booktitle = {Proc. ACM Web3D},
    month = {September},
    year = {2025},
    abstract = { We introduce PanoStyleVR, an immersive web-based framework for analyzing, ranking, and interactively applying style similarities within panoramic indoor scenes, enabling stereoscopic virtual exploration and photorealistic style adaptation. A key innovation of our system is a fully immersive \emph{WebXR} interface, allowing users wearing head-mounted displays to navigate indoor environments in stereo and apply new styles in real time. Style suggestions are visualized through floating thumbnails rendered in the VR space; selecting a style triggers photorealistic transfer on the current room view and updates the immersive stereo representation. This interactive pipeline is powered by two integrated neural components: (1) a geometry-aware and shading-independent GAN-based framework for semantic style transfer on albedo-reflectance representations; and (2) a gated architecture that synthesizes omnidirectional stereoscopic views from a single 360$^\circ$ panorama for realistic depth-aware exploration. Our system enables cosine-similarity-based style ranking, t-SNE-driven dimensionality reduction, and GMM-based clustering over large-scale panoramic datasets. These components support an immersive recommendation mechanism that connects stylistic analysis with interactive editing. Experimental evaluations on the Structured3D dataset demonstrate strong alignment between perceptual similarity and our proposed metric, and effective grouping of panoramas based on latent style representations. },
    note = {To appear},
    url = {http://vic.crs4.it/vic/cgi-bin/bib-page.cgi?id='Tukur:2025:PSS'},
}