YOLO Vision — Real-Time Object Detection

A FastAPI backend + polished web frontend for real-time object detection using YOLOv8.

Features

• 🖼 Image detection — upload any image, get annotated result with bounding boxes & confidence scores
• 🎞 Video detection — upload a video clip, download fully annotated MP4
• 📡 Live streaming — MJPEG stream with real-time detections from:
  • YouTube URLs (via yt-dlp)
  • Webcam (index 0)
  • Local video files
• Adjustable confidence threshold
• Per-class detection counts + confidence colour coding

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Run the server

cd yolo-app
python main.py
# or
uvicorn main:app --host 0.0.0.0 --port 8000

3. Open the UI

Navigate to http://localhost:8000

The YOLOv8-nano model (yolov8n.pt) is downloaded automatically on first run (~6 MB).

API Endpoints

Method	Path	Description
POST	/detect/image	Detect objects in an uploaded image
POST	/detect/video	Process a video and return annotated MP4
POST	/stream/start	Start a live detection stream
GET	/stream/mjpeg/{job_id}	MJPEG stream of annotated frames
GET	/stream/detections/{job_id}	Poll current detections as JSON
DELETE	/stream/{job_id}	Stop a running stream
GET	/health	Health check

Form Fields

POST /detect/image

• file — image file (multipart)
• conf — confidence threshold (0.0–1.0, default 0.3)

POST /detect/video

• file — video file (multipart)
• conf — confidence threshold
• max_frames — maximum frames to process (default 300)

POST /stream/start

• source_type — youtube | webcam | video_file
• url — YouTube URL (when source_type=youtube)
• file — video file (when source_type=video_file)
• conf — confidence threshold

Detection Response (X-Detections header on /detect/image)

[
  { "label": "person", "confidence": 0.923, "bbox": [x1, y1, x2, y2] },
  { "label": "car",    "confidence": 0.847, "bbox": [x1, y1, x2, y2] }
]

Model

Uses YOLOv8 nano by default (fast, ~6 MB). To switch to a larger model, change MODEL_NAME in main.py:

• yolov8n.pt — nano (fastest)
• yolov8s.pt — small
• yolov8m.pt — medium
• yolov8l.pt — large
• yolov8x.pt — extra-large (most accurate)

Architecture

Browser
  │
  ├─ POST /detect/image  ──► YOLO inference ──► Annotated JPEG response
  │
  ├─ POST /detect/video  ──► Frame loop ──► Annotated MP4 download
  │
  └─ POST /stream/start
       │
       ├─ GET /stream/mjpeg/{id}  ──► Multipart MJPEG (browser <img>)
       └─ GET /stream/detections/{id}  ──► JSON polled every 100ms