khornlund/severstal-steel-defect-detection

Kaggle Segmentation Challenge

40
/ 100
Emerging

Implements an ensemble of EfficientNet and SE-ResNeXt encoders paired with FPN/U-Net decoders using the segmentation_models.pytorch framework to detect and classify four types of surface defects on steel sheets. Training leverages mixed loss (0.6 BCE + 0.4 Dice), pseudo-labeling with balanced sampling, and grayscale input adaptation via weight recycling to achieve diverse model predictions. Achieves 0.91023 private leaderboard score through careful hyperparameter tuning, encoder/decoder architecture exploration, and image augmentation via Albumentations.

263 stars. No commits in the last 6 months.

No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 8 / 25
Community 22 / 25

How are scores calculated?

Stars

263

Forks

58

Language

Jupyter Notebook

License

Last pushed

Oct 07, 2020

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/khornlund/severstal-steel-defect-detection"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.