Computer Vision Object Detection App

I’m building a small AI-powered application whose core job is clear and focused: detect people in images or video using computer-vision object detection. The end goal is a lightweight, reliable tool that takes a file or live stream, runs inference, and returns bounding-box coordinates and confidence scores.

Here’s what matters most to me:

• A proven object-detection model (YOLOv8, Faster R-CNN, or a similar approach in PyTorch, TensorFlow, or OpenCV) fine-tuned if needed so that “person” detection is consistently accurate in varied lighting and crowd density.
• A simple interface—CLI, minimal web page, or desktop window is fine—where I can upload media and immediately see results overlaid or exported as JSON.
• Clean, documented code plus a short README covering setup, dependencies, and how to swap in a different dataset later.

Acceptance criteria
• ≥90 % precision and recall on a small validation set I’ll provide.
• Inference time under one second per 1080p frame on a mid-range GPU (e.g., RTX 3060) or equivalent CPU batch throughput.
• All project files delivered via Git, with reproducible environment specs (requirements.txt, environment.yml, or Dockerfile).

If you have a preferred framework or optimisation tricks—TensorRT, ONNX, quantisation—feel free to steer the build that way, as long as the accuracy and speed targets hold. I look forward to seeing what you can put together.