YOLOv8 Object Detection

Simple, real-time object detection using Ultralytics YOLOv8 and OpenCV.

This repo gives you a single script (detect.py) that can:

• run YOLOv8 on your webcam, images, or videos
• overlay bounding boxes, FPS, and per-class counts
• optionally save annotated output to disk

---

Features

• 🎥 Live webcam detection (--source 0 by default)
• 🖼️ Image & video file support
• 📊 Per-class object counts overlay (e.g. person: 3, car: 1)
• ⚡ FPS overlay showing inference speed
• 💾 Optional saving of annotated images/videos (--save, --out, --fps)
• 🧠 YOLOv8 model selection via --model (default: yolov8n.pt)
• 🧪 Verbose logging for debugging (--verbose)

---

Requirements

• Python: 3.8 – 3.12
  (avoid 3.13 for now, some wheels may be missing)
• OS: Windows, Linux, or macOS

Python dependencies (in requirements.txt):

ultralytics>=8.2.0
opencv-python>=4.8.0
numpy>=1.24.0

Install them with:

pip install -r requirements.txt

---

Model weights (yolov8n.pt)

By default the script uses:

--model yolov8n.pt

If yolov8n.pt is not found locally, Ultralytics will automatically download it on first run.

You can also:

• place yolov8n.pt in the project folder, or
• use any other YOLOv8 model, e.g. --model yolov8s.pt or a custom .pt file.

---

Quick start

1. Clone / download the repo

git clone <your-repo-url>
cd <your-repo-folder>

2. Install dependencies

pip install -r requirements.txt

3. Run with your webcam

python detect.py --source 0 --show

• 0 = default webcam
• add --verbose if you want detailed logs

4. Run on an image

python detect.py --source path/to/image.jpg --show --save

• --show → opens a window with detections
• --save → writes image_det.jpg next to the original

5. Run on a video file

python detect.py --source path/to/video.mp4 --show --save --fps 30

• --fps sets the output video FPS (for the saved file)

---

start.bat (Windows shortcut)

For convenience, the repo includes:

python detect.py --source 0 --show --verbose

You can:

• double-click start.bat in Explorer to start webcam detection with verbose logging
• or edit it if you want a different default model/source.

---

Command-line options

detect.py supports these arguments:

Option	Default	Description
--source	"0"	Input source: webcam index (0, 1, …), image path, video path, or stream URL (RTSP/HTTP).
--model	yolov8n.pt	YOLOv8 model weights file. Can be any .pt compatible with Ultralytics.
--conf	0.25	Confidence threshold (0–1). Higher = fewer, more confident detections.
--imgsz	640	Inference image size (pixels).
--device	auto	Device string, e.g. cpu, cuda:0. If omitted, Ultralytics auto-selects.
--show	False	Show a window with live detections.
--save	False	Save annotated image/video to disk.
--out	None	Output path for --save. If omitted, a default name is chosen.
--fps	0	Target FPS for saved video; 0 = use source FPS (or 30 if unknown).
--verbose	False	Print detailed startup and per-frame logs.

---

How it works (short version)

• Dependencies (opencv-python, ultralytics, numpy) are imported lazily in ensure_deps() so startup is fast.

• The script decides whether --source is:

  • a webcam index (like "0")
  • a file path (image/video)
  • or a generic stream URL

• For images:

  • runs model.predict(...)
  • draws boxes + class labels
  • overlays per-class counts
  • optionally saves the annotated image

• For video/streams:

  • reads frames in a loop
  • runs inference per frame
  • overlays boxes, FPS, and object counts
  • optionally shows a window and/or saves to a video file