LAÇİN

Brief

Serhan captains a five-person team building a YOLOv11 aerial detection system. He owns the training pipeline on a dataset fusing VisDrone, UAVDT, Stanford Drone, DOTAv2 and the team's own UAP/UAI labels, plus a custom augmentation stage that injects Gaussian blur and synthetic dead pixels so the model meets sensor failure before the flight line does.

What he built

• 01YOLOv11 training pipeline over a 116K-image dataset, reaching mAP@50 = 0.655
• 02Custom augmentation stage: Gaussian blur plus synthetic dead pixels
• 03Data augmentation and label visualizer GUI in CustomTkinter and OpenCV
• 04JSONtoYOLO converter mapping Supervisely exports into YOLOv11 structure
• 05JSON API server layer the rest of the team integrates against

Stack

• YOLOv11
• Python
• OpenCV
• SAHI
• CustomTkinter
• PyInstaller