Cristian Llull, Ivan Sipiran, Francisca Gil, Nelson Baloian, Anthony Belessis, Vlassis Fotis, George Ioannakis, Efthymios Koukoulis, Konstantinos Moustakas, Sotiris Papatheodorou, Ioannis Pratikakis, Ioannis Romanelis, Kriti Singh, Eleftheria Solomonidi, Christoforos Vlachos, Vaia Zachari
The geometric evaluation of 3D reconstruction techniques, spanning from classical structure-from-motion (SfM) to contemporary neural implicit representations and Gaussian splatting, remains a challenge. In cultural heritage applications, the preservation of high-frequency surface details, such as tool marks, inscriptions and reliefs, is often of greater significance than global shape approximation. Traditional metrics, notably the Chamfer Distance (CD), are frequently inadequate as their reliance on global error averaging masks the loss of critical micro-topography. To address this, in this Shape Retrieval Challenge (SHREC) track we introduce a novel benchmarking framework comprising: (1) a controlled, procedurally generated synthetic dataset of Khachkar-inspired stone reliefs, fully reproducible; and (2) the Feature-Aware Chamfer Distance (FACD), a feature-aware distance function designed to prioritize structurally and significantly geometric features. Three teams each submitted one run, and the organizers contributed five baseline reconstructions, for a total of eight configurations evaluated under a common protocol. Our evaluation reveals that standard CD frequently misranks models by favoring global smoothness over detail preservation. By providing more archaeologically plausible assessments, our FACD reveals a persistent fidelity gap, showing that current state-of-the-art methods still lack the nuance required for feature optimization.
Llull, C, Sipiran, I, Gil, F, Baloian, N, Belessis, A, Fotis, V, Ioannakis, G, Koukoulis, E, Moustakas, K, Papatheodorou, S, Pratikakis, I, Romanelis, I, Singh, K, Solomonidi, E, Vlachos, C & Zachari, V 2026, 'Shrec 2026: Reconstruction of high-frequency geometry in synthetic heritage', Computers & Graphics-uk, vol. 139, 104726. https://doi.org/10.1016/j.cag.2026.104726
Llull, C., Sipiran, I., Gil, F., Baloian, N., Belessis, A., Fotis, V., Ioannakis, G., Koukoulis, E., Moustakas, K., Papatheodorou, S., Pratikakis, I., Romanelis, I., Singh, K., Solomonidi, E., Vlachos, C., & Zachari, V. (2026). Shrec 2026: Reconstruction of high-frequency geometry in synthetic heritage. Computers & Graphics-uk, 139, Article 104726. https://doi.org/10.1016/j.cag.2026.104726
@article{bfa407c44162415fba05382c4c725e25,
title = "Shrec 2026: Reconstruction of high-frequency geometry in synthetic heritage",
abstract = "The geometric evaluation of 3D reconstruction techniques, spanning from classical structure-from-motion (SfM) to contemporary neural implicit representations and Gaussian splatting, remains a challenge. In cultural heritage applications, the preservation of high-frequency surface details, such as tool marks, inscriptions and reliefs, is often of greater significance than global shape approximation. Traditional metrics, notably the Chamfer Distance (CD), are frequently inadequate as their reliance on global error averaging masks the loss of critical micro-topography. To address this, in this Shape Retrieval Challenge (SHREC) track we introduce a novel benchmarking framework comprising: (1) a controlled, procedurally generated synthetic dataset of Khachkar-inspired stone reliefs, fully reproducible; and (2) the Feature-Aware Chamfer Distance (FACD), a feature-aware distance function designed to prioritize structurally and significantly geometric features. Three teams each submitted one run, and the organizers contributed five baseline reconstructions, for a total of eight configurations evaluated under a common protocol. Our evaluation reveals that standard CD frequently misranks models by favoring global smoothness over detail preservation. By providing more archaeologically plausible assessments, our FACD reveals a persistent fidelity gap, showing that current state-of-the-art methods still lack the nuance required for feature optimization.",
keywords = "3D reconstruction, 3D shape retrieval, Cultural heritage",
author = "Cristian Llull and Ivan Sipiran and Francisca Gil and Nelson Baloian and Anthony Belessis and Vlassis Fotis and George Ioannakis and Efthymios Koukoulis and Konstantinos Moustakas and Sotiris Papatheodorou and Ioannis Pratikakis and Ioannis Romanelis and Kriti Singh and Eleftheria Solomonidi and Christoforos Vlachos and Vaia Zachari",
year = "2026",
month = oct,
doi = "10.1016/j.cag.2026.104726",
language = "English",
volume = "139",
journal = "Computers \& Graphics-uk",
issn = "0097-8493",
publisher = "Elsevier",
}