bloombio/phytoplankton-microscopy

Bloombio Phytoplankton Microscopy Dataset

The Bloombio Phytoplankton Microscopy Dataset is a curated, citable marine science dataset consisting of 43 phytoplankton species across 10,433 high-resolution light microscopy images with bounding box annotations.

Developed as part of the Bloombio Marine Intelligence Platform — a platform that equips scientists with an autonomous AI agent, compressing sampling setup, multi-modal species identification, and environmental risk assessment from days to seconds.

---

Dataset Summary

Attribute	Value
Total Images	10,433 PNG files
Species Classes	43 phytoplankton taxonomies
Annotations	Bounding boxes (Pascal VOC XML source, normalized YOLO format in HF)
Image Resolution	~3584 × 2748 px (High-resolution light microscopy)
Splits	70% Train, 15% Validation, 15% Test (Stratified)
License	Creative Commons Attribution 4.0 International (CC BY 4.0)

---

How to Use

Load directly in Python using the Hugging Face datasets library:

1. Classification Configuration (classification)

Contains 512×512 pre-cropped organism patches (white-padded square preserving aspect ratio with 5% border padding). Ideal for training vision classifiers (ViT, ResNet, EfficientNet, OpenCLIP).

from datasets import load_dataset

dataset = load_dataset("bloombio/phytoplankton-microscopy", "classification")
print(dataset)
# Returns: DatasetDict({ train: 7282, validation: 1566, test: 1584 })

# Access an example
sample = dataset["train"][0]
image = sample["image"]      # PIL Image (512x512)
label = sample["label"]      # Class integer ID
source = sample["source_file"]

2. Object Detection Configuration (detection)

Contains full-resolution microscopy slides with normalized YOLO bounding boxes ([x_center, y_center, width, height]). Ideal for training object detectors (YOLO, DETR).

from datasets import load_dataset

dataset = load_dataset("bloombio/phytoplankton-microscopy", "detection")

# Access an example
sample = dataset["train"][0]
image = sample["image"]      # Full resolution PIL Image
bbox = sample["bbox"]        # [x_center, y_center, width, height] normalized
label = sample["label"]      # Class integer ID

---

Class Distribution (43 Taxonomies)

Species	Count	Species	Count
Pediastrum	920	Coscinodiscus	170
Rhizosolenia	863	Gyrosigma	168
Scenedesmus	711	Aphanocapsa	167
Aulacoseira	657	Diploneis	154
Chaetoceros	519	Nitzschia closterium	125
Navicula	489	Actinastrum	119
Cosmarium	480	Sphaerocystis sp	106
Arthrospira platensis	467	Phacus	82
Staurastrum	446	Eucampia	75
Asterionellopsis glacialis	442	Cerataulina pelagica	70
Merismopedia	419	Isthmochloron	66
Guinardia flaccida	402	Tetraeadron	58
Nitzschia acicularis	314	Coelastrum	47
Micrasterias	265	Ulnaria acus	41
Melosira	254	Synedra acus	41
Ulnaria ulna	251	Prorocentrum gracile	40
Anabaena	245	Sorastrum americanum	37
Thalassionema nitzschiodes	221	Dictyosphaerium sp	31
Closterium gracile	206	Bacteriastrum	31
Peridinium	192	Pinnularia	29
		Westella botryoides	6
		Pseudo-nitzschia	4
		Xanthadium faciculatum	2

Note: Rare classes with low counts are preserved to support research in long-tail/few-shot taxonomy identification and will be augmented in future versions.

---

Citation Information

If you use this dataset in your research or project, please cite:

@dataset{bloombio-phytoplankton-v1.0.0,
  title     = {Bloombio Phytoplankton Microscopy Dataset},
  author    = {Baidoo, Fiifi and Amamoo, Phyllis},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/bloombio/phytoplankton-microscopy},
  license   = {CC BY 4.0}
}

---

License

This dataset is distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.