view article
Figure 7
Quantitative benchmarking and performance evaluation of YOLO architectures for feature tracking in high-speed X-ray images. (a) Visual comparison illustrating the high spatial overlap between the manually annotated ground truth (green) and the YOLO prediction (red). (b-1 to b-3) Comparative analysis of YOLOv8 against YOLOv12. Results demonstrate that YOLOv8 requires lower peak GPU VRAM, compared to YOLOv12 (b-1) and demonstrates faster training times per epoch (b-2), and achieving equivalent segmentation accuracy (b-3), thereby validating YOLOv8's integration into SMAXI's object tracking pipeline.

Journal logoJOURNAL OF
SYNCHROTRON
RADIATION
ISSN: 1600-5775
Volume 33| Part 5| September 2026|
https://doi.org/10.1107/S1600577526007745