Pseudo-nitzschia is a genus of harmful algal bloom (HAB) forming diatoms. Cells grow in stepped chains, making individual cell counting challenging. This dataset provides polygon annotations of individual cells within chain colonies, enabling automated cell counting via instance segmentation.

Images were collected by IFCB104 at the Santa Cruz Municipal Wharf, Monterey Bay, California.

Dataset Structure

data/
  images/    # 302 grayscale PNG images of Pseudo-nitzschia chains
  labels/    # 301 YOLO polygon segmentation label files
train.txt    # 180 images
val.txt      # 60 images
test.txt     # 61 images
instances_default.json  # COCO-format annotations (original CVAT export)

Annotation Format

Labels are in YOLO polygon segmentation format:

class_id x1 y1 x2 y2 x3 y3 ... xn yn

class_id: always 0 (single class: Pseudo_nitzschia_cell)
Coordinates are normalized to [0, 1]
Each line represents one cell polygon
Multiple lines per file = multiple cells in the image

Class Information

ID	Name	Description
0	`Pseudo_nitzschia_cell`	Individual cell within a chain colony

Statistics

Total images: 302
Annotated images: 301 (with at least one cell polygon)
Train/Val/Test split: 180 / 60 / 61
Image format: Grayscale PNG (variable dimensions, typically ~600x50 px)
Source instrument: IFCB104 (Imaging FlowCytobot)

Usage with YOLO

# cell_seg.yaml
path: ./data
train: ../train.txt
val: ../val.txt
test: ../test.txt
nc: 1
names:
  0: Pseudo_nitzschia_cell

yolo segment train data=cell_seg.yaml model=yolo11s-seg.pt epochs=100

Source

Instrument: IFCB104 - Santa Cruz Municipal Wharf
Annotation tool: CVAT
Associated model: patcdaniel/pseudo-nitzschia-yolo11s-seg

Citation

If you use this dataset, please cite:

@dataset{pseudo_nitzschia_cell_seg,
  author = {Daniel, Patrick C.},
  title = {Pseudo-nitzschia Cell Segmentation Dataset},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/patcdaniel/pseudo-nitzschia-cell-segmentation}
}