Challenges in Transferring Satellite Representation Learning to Pathology

gabriberton · x · 2026-08-25

Discussing transfer learning in pathology, the author notes that natural image models (like DINOv2) transfer better to satellite imagery than to pathology due to significant domain differences. Additionally, satellite representation learning often leverages multi-source data (street view, GPS, elevation) for contrastive learning, a strategy not easily applicable in pathology.

Original post →

More from Research

Research channel →