Computer vision that finds steel-surface defects at pixel level (UNet++ / EfficientNet), not just pass or fail.
Problem
Finding surface defects in photos by hand is slow and inconsistent.
Founder’s role
Dmitri trained a segmentation model (UNet++, EfficientNet) that marks defects at pixel level.
Result
Steel-surface defects are found and shown precisely, not just "defective or ok".
Problem
Image-level pass/fail does not tell an inspector where a defect is. Small, low-contrast marks on textured steel are easy for a classifier to miss.
Solution
A segmentation pipeline: mask training, IoU/Dice evaluation, FastAPI inference and a browser overlay so quality teams can inspect predicted defect regions.
Result
Defects are shown precisely on the image, not just with a label; the model and code are public and repeatably deployable.
