We present PhyEddy-Net, a daily global mesoscale eddy detection dataset spanning the ice-free ocean between 60° S and 60° N for the period 2010–2021. The dataset is produced by a multimodal physics–data dual-driven deep-learning segmentation framework that integrates sea level anomaly (SLA), geostrophic relative vorticity, and sea surface temperature anomaly (SSTA). Each daily record provides the eddy centroid, polarity (cyclonic/anticyclonic), area, equivalent-circle radius, detection probability, and ocean basin. The full catalogue comprises approximately 36.7 million detections, corresponding to roughly 3.06 million per year. Independent leave-one-Argo-profile-out validation of the released PhyEddy-Net product against 5,176 clear Argo profiles over the overlapping 2010–2012 period yields a hit rate of 87.1 % and a type agreement of 92.1 % following the production LOO polarity correction. The released product is also compared with META4, Faghmous et al. (2015), and the CASEarth V3.0 SLA atlas using identical profiles and matching criteria. The dataset is intended for studies of mesoscale eddy occurrence, spatial distribution, seasonal-to-interannual variability, and ocean model validation, and is available from Mendeley Data (Qiu et al., 2026) at <a href="https://doi.org/10.17632/rh6wtx3hyr.2" target="_blank" rel="noopener">https://doi.org/10.17632/rh6wtx3hyr.2</a>.