CROWN: 10M cytology images beat 100x-larger pathology foundation models
bravo_abad · x · 2026-09-19
Zheng et al. trained CROWN, a pathology foundation model, on over 10 million cytology images via self-supervised learning — no manual labels needed — targeting the visual language of isolated cells and small clusters, distinct from tissue architecture in histopathology.
On 12 public cytology datasets, CROWN reaches 0.886 mean accuracy, beating UNI (0.874, pretrained on 100M+ histopathology patches) and Virchow (0.848, a 632M-parameter model trained on 1.5M whole-slide images).
Takeaway: domain-matched pretraining data can matter more than scale for scientific foundation models.
More from Research
- Meta's Atomic Creator Tokens beat Semantic IDs: +19.8pt Recall@15 and 2.7x lower latency — _reachsumit · 2026-11-04
- HiGenTrans transfers LLM knowledge into generative recommenders at input, representation and decision levels — _reachsumit · 2026-11-04
- SA-OPSD distills self-teacher knowledge at beam-search pruning boundaries for generative recommendation — _reachsumit · 2026-11-04
- MV-Thinker: Multi-View Thinking with Adaptive Generation for Generative Recommendation — _reachsumit · 2026-11-04
- ByteDance details USDB storage system: 85.8% lower write latency for recommender sequences — _reachsumit · 2026-11-04
- Kuaishou's UAGR injects utility signals into semantic codes and decoding for generative recommendation — _reachsumit · 2026-11-04