Skip to content

Latest commit

 

History

History
2 lines (2 loc) · 360 Bytes

File metadata and controls

2 lines (2 loc) · 360 Bytes

severstal-steel-defect-detection

This project was to improve algorithms for localizing and classifying surface defects on a steel sheet. Models were built with sample base CNN to dense NET to VGG for classification, and with different UNET combined with data augmentation to further boost the performance for segmentation. Algorithms are tested on Kaggle.