I will train a custom computer vision model for defect detection or counting

Hands-on experience in real industrial computer vision projects, object detection and defect classification models trained on local GPU hardware with COCO-format annotation, contour-based object counting, and separating false positives from real defects on a live production line.

What's included:

• Custom-trained model for your specific defect or counting task
• COCO-format annotation of your images (done by me, or I'll guide you on preparing it)
• Accuracy validation on a held-out test set before delivery
• Inference code ready to run, so you can test it on new images yourself
• Documentation explaining how the model works and how to use it

My approach starts with a feasibility test on a small sample of your images, so you know the model's real accuracy before committing to a bigger package. I deliver a trained model and working inference code; integrating it into your production line or existing software is done together, based on your setup.