draineye/draineye-materials

DrainEye — Drainage Material Classification Dataset

An image classification dataset for identifying pipe and drainage materials, assembled from public Roboflow and MINC-2500 sources.

Classes

• asbestos — asbestos cement pipes and surfaces
• hdpe — high-density polyethylene (HDPE/ПНД) plastic pipes
• ceramic — ceramic surfaces and pipes
• stone — stone and slag-like surfaces

Dataset Structure

Split	hdpe	ceramic	stone	asbestos	Total
train	384	1750	1750	1200	5084
val	109	500	500	343	1452
test	56	250	250	172	728
Total	549	2500	2500	1715	7264

Split: 70% train / 20% val / 10% test, random.seed(42).

Class imbalance present — recommend using class_weight during training.

Sources

• Asbestos91 INAIL — Roboflow Universe (CC BY 4.0)
• Plastic HDPE — Roboflow Universe (CC BY 4.0)
• MINC-2500 — Materials in Context Database, Bell et al., CVPR 2015

Intended Use

Training a mobile drainage material classifier (Flutter + TFLite) as part of the DrainEye inspection app.

Dataset consists of general-purpose material images, not in-pipe photography.
Expect distribution shift when deployed on real drainage footage.

Loading

from datasets import load_dataset
dataset = load_dataset("draineye/draineye-materials")