Cellpose
Cellpose is a deep learning model for cell segmentation. It detects and outlines individual cells in microscopy images, supporting multiple cell types and automatic diameter estimation.Input formats
.png, .jpg, .jpeg, .tiff, .bmp, .webp
Model variants
| Variant | Description |
|---|---|
| cyto3 (default) | Latest cytoplasm model — best general-purpose accuracy |
| cyto2 | Second-generation cytoplasm model |
| cyto | Original cytoplasm model |
| nuclei | Nuclei-only detection |
Parameters
| Parameter | Range | Default | Description |
|---|---|---|---|
| Model type | cyto/cyto2/cyto3/nuclei | cyto3 | Which Cellpose model to use |
| Diameter | 0–500 px | 0 (auto) | Expected cell diameter. Set to 0 for auto-detection. |
| Flow threshold | 0–1 | 0.4 | Maximum flow error per mask (advanced) |
| Cell probability threshold | -6 to +6 | 0.0 | Cell probability cutoff (advanced) |
| Channels | [int, int] | [0, 0] | Cytoplasm and nucleus channel indices (advanced) |
Outputs
| Output | Format | Description |
|---|---|---|
| Segmentation mask | PNG | Combined instance segmentation mask |
| Polygons | JSON (GeoJSON) | Vector outlines of each detected cell |
Presets
| Preset | Description |
|---|---|
| Cell Segmentation | cyto3 model with auto-diameter |
| Nuclei Only | nuclei model for nuclear detection |
Compute requirements
| Resource | Requirement |
|---|---|
| GPU | Required — min 8 GB VRAM, 16 GB recommended |
| Duration | ~30 seconds per image |