FDSML: Formula-Driven Metric Learning Boosts Image Retrieval Without Real Images

A Japanese team's FDSML paper, accepted by WACV 2027, uses FractalDB generation parameters as supervision for metric learning, improving Stanford Cars retrieval Recall@1 by 22.2 points without any real-image pretraining.

2026-10-31 ~ 2026-10-31 · 2 related posts

• FDSML (WACV 2027): fractal-parameter metric learning lifts image retrieval up to +22pt — HirokatuKataoka · 2026-10-31
• FDSML pretrains without real images, lifting image retrieval Recall@1 by 22.2 points — HirokatuKataoka · 2026-10-31