Assess whether synthetic data can fully replace real-world labels

Ascertain whether synthetic egocentric hand–object interaction datasets can fully substitute for labeled real data without degrading performance in hand–object interaction detection.

Background

A central practical question is whether synthetic data can obviate the need for costly real-world annotation. If synthetic-only training achieves comparable performance, it would dramatically reduce labeling effort.

The paper benchmarks synthetic-only models against models trained with real annotations to evaluate this possibility across multiple datasets.

References

As a result, several key open questions still need to be addressed: 1) How large is the gap between synthetic and real data? 2) What are its main causes? 3) How can it be minimized? 4) Can synthetic data fully replace real-world data?

— Leveraging Synthetic Data for Enhancing Egocentric Hand-Object Interaction Detection  (2603.29733 - Leonardi et al., 31 Mar 2026) in Section 1 (Introduction)