2026-10-06T05:20:20Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34070919 2026-10-05T17:11:53Z category_28879 category_28885 category_29161 item_type_3 month_year_10_2026

Code, split files, out-of-fold predictions and latency measurements for "Coil leakage, training variance and FLOP proxies in the evaluation of real-time steel surface defect detectors" Zehao Li (25144737) Computer vision Image processing Deep learning steel surface defect detection NEU-DET GC10-DET data leakage cross-validation inference latency YOLO26 TensorRT <p dir="ltr">Supplementary material for the article "Coil leakage, training variance and FLOP proxies in the evaluation of real-time steel surface defect detectors" (submitted to the Journal of Real-Time Image Processing).</p><p dir="ltr">The study re-evaluates real-time steel surface defect detectors on NEU-DET and GC10-DET with five-fold cross-validation, coil-disjoint folds for GC10-DET, repeated training, paired comparisons and end-to-end latency measured with four inference runtimes on three platforms (NVIDIA RTX 5090, AMD Ryzen 7 9850X3D CPU, Apple M3 Pro).</p> 2026-10-05T17:11:53Z Dataset Dataset 10.6084/m9.figshare.34070919.v1 https://figshare.com/articles/dataset/Code_split_files_out-of-fold_predictions_and_latency_measurements_for_Coil_leakage_training_variance_and_FLOP_proxies_in_the_evaluation_of_real-time_steel_surface_defect_detectors_/34070919 CC BY 4.0