Cross-view agreement rescoring improves tiled small-object detection without retraining: result grid, per-detection dumps and code

Résumé fourni par la source

Supplementary data and code for the manuscript "Cross-view agreement rescoring improves tiled small-object detection without retraining" (submitted to Scientific Reports, 2026). Contents: (1) the full 110-configuration result grid (detector x benchmark x view generator x merge rule) as CSV/XLSX; (2) per-detection dumps (Parquet) for every configuration, from which all tables and figures are computed; (3) the numbers behind every table and figure; (4) the archived experimental pipeline (view generation, agreement computation, rescoring, evaluation and table generation), released under the AGPL-3.0 licence. Data files are released under CC BY 4.0; the code archive retains its AGPL-3.0 licence. The benchmark images (VisDrone2019-DET, DOTA-v1.5, AI-TOD-v2) are not redistributed and must be obtained from their original distributors.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.