You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Microplastic microscopy dataset

This access-controlled dataset contains curated microscopy images and instance-level masks used to train and evaluate the accompanying particle segmentation and morphology-classification pipeline.

Dataset summary

  • 241 microscopy images
  • 339 reviewed particle instances
  • 257 instances included in detector training and evaluation
  • 334 instances with PS, PP, or PET morphology labels
    • PS: 105
    • PP: 179
    • PET: 50

The five instances without a morphology label are retained for particle detection but are not used to train the Logistic Regression classifier.

Files

gt_instances.csv
images/{source_image_id}.png
masks/{source_image_id}/particleN_mask.npz

gt_instances.csv contains the instance identifier, source image, group, five-fold split, image and mask paths, bounding box, mask area, morphology label, and detector-inclusion flag.

Masks are stored as SciPy sparse arrays:

import numpy as np
import scipy.sparse

sparse_mask = scipy.sparse.load_npz(mask_file)
mask = np.asarray(sparse_mask.todense()).reshape(sparse_mask.shape[-2:]).astype(bool)

Cross-validation split

The fold assignment was generated with grouped stratification and a fixed random seed. Instances from the same source image remain in the same fold to avoid image-level leakage. The repository code uses one fold for testing, the next fold for validation, and the remaining three folds for training.

Labels and intended use

The PS, PP, and PET labels are morphology-based experimental classes under the sample-preparation and acquisition conditions used in this study. They support the reported shape-based classification experiments; they are not a substitute for independent chemical polymer identification.

The paired code and checkpoints are available at:

Limitations

  • The class distribution is imbalanced, particularly for PET.
  • Optical morphology is affected by preparation, imaging, overlap, and image quality.
  • Chemical confirmation requires FTIR, Raman spectroscopy, or an equivalent analytical method.
  • Access does not imply permission to redistribute the images. Follow the repository owner's data-use terms.
Downloads last month
576