EUVIMEDEUVIMED entdecken
Crossref

Conditioning GMFlow with a Gated Cross-Correlation Prior for Particle Image Velocimetry

Aarnav Trivedi, Talya H. Stehley

Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 675-679

DOI: 10.1609/aaaiss.v10i1.43474 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Aarnav Trivedi und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Particle image velocimetry (PIV) estimates dense displacement fields from pairs of tracer images. Classical cross-correlation provides local displacement estimates and peak diagnostics, whereas neural optical-flow models can use wider spatial context. We introduce ANCHOR-PIV, which adds a clipped,validity-masked cross-correlation prior to GMFlow’s pre-softmax matching scores. A small gate combines eight input features to control the local prior strength, and all backbone weights remain trainable. We compare ANCHOR-PIV to GMFlow, using three training seeds for each model. EPE averaged equally across the nine flow conditions (0.2791 px and 0.2775px for GMFlow and ANCHOR-PIV, respectively) was computed over the 450 benchmark pairs. The absolute difference was -0.0016 px (paired 95% CI [-0.0029, +0.0001]; p = 0.0625), a 0.58% relative reduction. This is below our project-specific 5% improvement target. The gate increased the neural parameters by 0.018%. Neural forward time was 0.025s/pair compared to 0.012s/pair for direct GMFlow. Forward cross-correlation took 0.075s/pair. A complete forward reverse analysis with error scoring took 0.247s/pair. Posthoc prediction intervals were severely overconservative. On this benchmark, the prior did not establish a meaningful accuracy improvement and increased computation.

Zitieren

APA

Trivedi, A., & Stehley, T. H. (2026). Conditioning GMFlow with a Gated Cross-Correlation Prior for Particle Image Velocimetry. Proceedings of the AAAI Symposium Series, 10(1), 675-679. https://doi.org/10.1609/aaaiss.v10i1.43474

Vancouver

Trivedi A, Stehley TH. Conditioning GMFlow with a Gated Cross-Correlation Prior for Particle Image Velocimetry. Proceedings of the AAAI Symposium Series. 2026;10(1):675-679. doi:10.1609/aaaiss.v10i1.43474

BibTeX
@article{Trivedi2026,
  author = {Trivedi, Aarnav and Stehley, Talya H.},
  title = {Conditioning GMFlow with a Gated Cross-Correlation Prior for Particle Image Velocimetry},
  journal = {Proceedings of the AAAI Symposium Series},
  year = {2026},
  volume = {10},
  number = {1},
  pages = {675--679},
  doi = {10.1609/aaaiss.v10i1.43474}
}
Kostenlos starten →