severstal

A collection of public datasets for surface metal defects detection task. Source, and roboflow (ready to use) links are included.

Steel surface defect detection using YOLOv8m on the Severstal dataset. Full pipeline: RLE-to-bbox conversion, image tiling, stratified splitting, custom balanced sampling, and Gradio deployment on HuggingFace Spaces.

Another aproach to a Severstal test task

116th place Kaggle Severstal Steel Defect Detection competition 2019

Kaggle competition for image segmentation of surface defects on steel sheets. Utilizing Convolutional Neural Networks (CNNs) to localize and classify defects, presenting a challenge due to subtle defect types and complex textures.