Industrial defect Image Classification Dataset

379 labeled images

Premium dataset

379+ labeled industrial defect images ready for AI and computer vision research. Delivered as an ImageFolder-style ZIP with train / val / test splits and class-labeled folders. Review the source license before use. Compatible with PyTorch, TensorFlow, Keras, and any classification pipeline.

Download dataset · 379 images Support the project

Need a production-grade version?

Our B2B team can deliver a licensed, commercial-ready version of this dataset — with custom labels, QA, and annotation formats.

Talk to our team

Browse a selection of samples from the Industrial defect image dataset.

Categories in this dataset

Chipped brick Moss covered brick Melted wire Fabric pilling texture Fabric hole Stained paper Water stained brick Bubbled paint Fabric tear Creased paper Cracked insulation Plastic injection molding defect Fabric color fading Eroded brick Overheated wire Chipped paint Water damaged paper Cracked paint Paint uneven coating Uneven brick wall Burnt edges Tangled cables Plastic deep scratch Cracked plastic Stained fabric Plastic surface scratch Fabric uneven weave Fabric frayed edge Exposed copper Fabric wrinkled surface Folded corner Fabric loose thread Broken mortar Dripped paint Loose brick Plastic melted spot Frayed wire Punched holes Discolored plastic Wrinkled paper Peeling paint Warped plastic Cracked brick Dirt in paint Ink smudges Plastic uneven texture Stretched fabric Burnt brick Cut wire Loose wire Scratched paint Discolored brick Torn paper Blistered surface Faded color Burnt wire end Corroded connection Air bubble Dented plastic

Data sources

Plasticdefect

What you get

Industrial defect classification dataset, ready to train

Clean folder layout

`train` / `val` / `test` sub-folders per class plus a `meta.json` with label list, split counts, and ImageNet normalization constants — ready for `torchvision.datasets.ImageFolder`.

Your choices, your ZIP

Pick output size, color mode, train/val/test split, and 20 augmentations. The ZIP is built for your exact setup.

Research-friendly

BibTeX citation per dataset and a Colab starter notebook. No signup required to download.

Framework-agnostic

PyTorch, TensorFlow, Ultralytics — pick any. Standard ImageFolder layout works out of the box.