Dataset Card for severstal steel defects This is a FiftyOne dataset with 18074 samples. Installation If you haven't already, install FiftyOne: Usage Dataset Details Dataset Description Severstal: Steel Defect Detection is a surface defect segmentation dataset released for a 2019 Kaggle competition by Severstal, a Russian steel and mining company. The task is to localize and classify defects on flat sheet steel from high frequency camera images. Each image may contain no defects, a single defect class, or multiple defect classes (ClassId 1 4). Annotations were originally provided as run length encoded (RLE) masks in train.csv. Curated by: Severstal (Kaggle contributors: Alexey Grishin, BorisV, iBardintsev, inversion, Oleg) Funded by: Severstal PAO Shared by: Severstal, via Kaggle (2019) Language(s) (NLP): N/A (image data; metadata in English) Dataset Sources Repository: https://www.kaggle.com/competitions/severstal steel defect detection Paper: None (competition dataset, no accompanying paper) Uses Direct Use Training and benchmarking semantic / instance segmentation models for surface defect detection on flat sheet steel. Multi label image classification (defect present / absent pe…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy