Uzair Shah, Giovanni Pintore, Muhammad Tukur, Zahoor Ahmad, Jens Schneider, Matteo Sgrenzaroli, GiorgioVassena, Pere-Pau Vázquez, Enrico Gobbetti, and Marco Agus
2026
Abstract
Monocular 360-degree depth estimation is fundamental to full-scene 3D understanding. Deep learning effectively infers depth from incomplete or ambiguous monocular cues by leveraging large-scale data priors. Holistic panoramic methods further exploit global context to achieve structurally consistent predictions. However, high memory and computational costs, together with limited high-resolution training data, restrict existing holistic 360-degree methods to inputs orders of magnitude smaller than those of modern 8K–60 MP cameras and head-mounted displays (HMDs). Post-hoc upsampling offers only partial improvement, failing to recover original detail, while current approaches that stitch depth from multiple limited-FoV views increase local accuracy but compromise global consistency. To overcome these limitations, we introduce FRED - Full-Resolution Equirectangular Depth - a unified framework combining holistic 360-degree structural recovery with high-fidelity perspective refinement. Our training-free pipeline first obtains a coarse panoramic prior from a pretrained model, then performs high-resolution inference on overlapping gnomonic projections uniformly distributed on a midpoint-icosahedral sphere. Local predictions are refined using a prompt-guided foundation depth model and fused via a low-complexity seam-robust spherical blending strategy that preserves the global structure and local details. Our approach scales linearly with resolution, maintaining fine spatial detail and enabling accurate 8K–60MP inference with limited computational demands.
Reference and download information
Uzair Shah, Giovanni Pintore, Muhammad Tukur, Zahoor Ahmad, Jens Schneider, Matteo Sgrenzaroli, GiorgioVassena, Pere-Pau Vázquez, Enrico Gobbetti, and Marco Agus. FRED: Full-Resolution Equirectangular Depth Estimation. In Spanish Computer Graphics Conference (CEIG), 2026. DOI: 10.2312/ceig.20261002.
Related multimedia productions
Bibtex citation record
@inproceedings{Shah:2026:FFE, author = {Uzair Shah and Giovanni Pintore and Muhammad Tukur and Zahoor Ahmad and Jens Schneider and Matteo Sgrenzaroli and GiorgioVassena and Pere-Pau Vázquez and Enrico Gobbetti and Marco Agus}, title = {{FRED}: Full-Resolution Equirectangular Depth Estimation}, booktitle = {Spanish Computer Graphics Conference (CEIG)}, year = {2026}, abstract = { Monocular 360-degree depth estimation is fundamental to full-scene 3D understanding. Deep learning effectively infers depth from incomplete or ambiguous monocular cues by leveraging large-scale data priors. Holistic panoramic methods further exploit global context to achieve structurally consistent predictions. However, high memory and computational costs, together with limited high-resolution training data, restrict existing holistic 360-degree methods to inputs orders of magnitude smaller than those of modern 8K–60 MP cameras and head-mounted displays (HMDs). Post-hoc upsampling offers only partial improvement, failing to recover original detail, while current approaches that stitch depth from multiple limited-FoV views increase local accuracy but compromise global consistency. To overcome these limitations, we introduce {FRED} --- Full-Resolution Equirectangular Depth --- a unified framework combining holistic 360-degree structural recovery with high-fidelity perspective refinement. Our training-free pipeline first obtains a coarse panoramic prior from a pretrained model, then performs high-resolution inference on overlapping gnomonic projections uniformly distributed on a midpoint-icosahedral sphere. Local predictions are refined using a prompt-guided foundation depth model and fused via a low-complexity seam-robust spherical blending strategy that preserves the global structure and local details. Our approach scales linearly with resolution, maintaining fine spatial detail and enabling accurate 8K–60MP inference with limited computational demands.}, doi = {10.2312/ceig.20261002}, url = {http://vic.crs4.it/vic/cgi-bin/bib-page.cgi?id='Shah:2026:FFE'}, }
The publications listed here are included as a means to ensure timely
dissemination of scholarly and technical work on a non-commercial basis.
Copyright and all rights therein are maintained by the authors or by
other copyright holders, notwithstanding that they have offered their works
here electronically. It is understood that all persons copying this
information will adhere to the terms and constraints invoked by each
author's copyright. These works may not be reposted without the
explicit permission of the copyright holder.
Please contact the authors if you are willing to republish this work in
a book, journal, on the Web or elsewhere. Thank you in advance.
All references in the main publication page are linked to a descriptive page
providing relevant bibliographic data and, possibly, a link to
the related document. Please refer to our main
publication repository page for a
page with direct links to documents.
+39 070 9250 1
+39 070 9250 216
(Secretary) katia
crs4.it
www.crs4.it/vic/
