Cellpose-SAM hits 0.163 error rate, beating human annotator disagreement on cell segmentation
RichardsonDx · x · 2026-11-10
- Cellpose-SAM, built on Meta's SAM vision model, cuts cell segmentation error to 0.163 — below the 25% disagreement between two experts tracing the same cells, and near the best achievable score
- Paper figures (67-image Cellpose test set) show it beats Cellpose, CellSAM, and Omnipose on error rate and average precision, segmenting an image in 1 second on an A100
- It generalizes to stains and microscopes it has never seen before
Related event: Cellpose-SAM Beats Inter-Expert Disagreement in Cell Segmentation(2 posts)→
More from Research
- Zero-dependency MCP hub fronts nine research repos as 14 tools without merging dependency trees — Emergency-Tutor-7 · 2026-11-11
- tts-bench round 3: 91 open TTS models and 28 paid APIs benchmarked on identical prompts — UkieTechie · 2026-11-11
- GraphRAG has a schema drift problem; Neo4j's ontology spec proposes a shared contract — JeremyCMorgan · 2026-11-11
- New philosophy paper: Moti Mizrahi replies to Nyholm on AI 'mental states' and longtermism — SvenNyholm · 2026-11-10
- UT Austin open-sources OopsieVerse, a damage-aware safety benchmark for household robots — DynamicWebPaige · 2026-11-10
- Scott Aaronson on AI solving centuries-old math problems, P vs NP debate — burny_tech · 2026-11-10