tanırım
// Computer Vision · Model Evaluation · In-Browser Inference
TANIRIM

I build models
that make sense
of images.

I evaluate models with tools I wrote myself and make them run in the browser, with nothing to install.

Numan Bekir ÖzçelikAI & Computer Vision Engineer

// Work

Projects

Aerial Object Detection

YOLO11s trained on VisDrone for object detection in drone footage. The same model was trained at 640 and 1280 input; both run side by side in the browser, with nothing to install.

Small-object AP, input 640 → 1280
0.111 → 0.223
Per image, Chrome WebGPU (640)
23 ms
  • YOLO11s
  • VisDrone
  • ONNX Runtime Web
  • WebGPU