Vison_Learning

VocabLearn

📚 VocabLearn

AI-Powered real-time vocabulary trainer — point your webcam at everyday objects and learn their English names instantly using state-of-the-art object detection.

VocabLearn uses YOLOv8 to detect objects from your webcam or uploaded images and builds an interactive vocabulary list for learning and review.

✨ Key Features

🎥 Real-time detection with a live webcam feed (Frame-by-frame YOLOv8 inference)
Stylish detection boxes with confidence scores
Support for multiple camera sources (built-in/external/USB)
📖 Automatic vocabulary tracking (word collection, counts, visual progress)
📊 Session statistics (words learned, total detections, most detected, session duration)
📸 Screenshot capture, gallery, and per-image downloads
📁 Image upload mode for offline/image-based detection
💾 Export vocabulary as JSON for use with flashcard tools
⚙️ Adjustable settings: confidence threshold, camera source, model selection

🧩 What it Detects

VocabLearn uses the YOLOv8 Nano model (yolov8n.pt) and recognizes the 80 object classes from the COCO dataset (everyday object categories). You can change the model to yolov8s, yolov8m, yolov8l, or yolov8x for different accuracy/speed trade-offs.

🚀 Quick Start

Requirements

Python 3.9 or higher
Webcam (for live detection) or images for upload mode
pip package manager

Clone the repository

git clone https://github.com/Gus-Fring666/Vison_Learning.git
cd Vison_Learning

Install dependencies

Install from requirements:

pip install -r requirements.txt

Or install manually:

pip install streamlit ultralytics opencv-python-headless numpy Pillow

Note: Use opencv-python instead of opencv-python-headless if you plan to run the OpenCV desktop app with native windows.

Model weights

The yolov8n.pt weights will be automatically downloaded by the Ultralytics library on first run. If you prefer, place a model file in the project root.

🖥️ Running the App

Streamlit (recommended):

streamlit run streamlit_app.py

OpenCV desktop app (original):

python app.py

Streamlit will open at http://localhost:8501 by default.

🎮 Usage Guide

Live Detection

Click “Start Detection” to enable the webcam.
Point the camera at everyday objects.
Watch detections appear and be added to your vocabulary list automatically.
Click “Stop Detection” to pause.

Image Upload Mode

Use “Upload Image” to analyze local images (JPG, PNG, BMP, WebP).
Results are shown with the same detection overlays and added to vocabulary.

Screenshots

Click “Screenshot” while the camera is running to save the current annotated frame to screenshots/.
View and download screenshots from the gallery.

Export Vocabulary

Click “Download Vocabulary (JSON)” to export detected words and counts.

⚙️ Configuration & Settings

MODEL_NAME: yolov8n.pt by default — change for better/worse speed/accuracy.
Confidence threshold: adjustable (10%–95%) — default recommended ~0.45.
Camera index: default 0. Use the Camera Source dropdown to switch.

Recommended presets:

High accuracy: yolov8l or yolov8x with higher confidence threshold (0.6–0.75).
Highest speed: yolov8n with moderate threshold (0.35–0.5).

🐛 Troubleshooting

Camera not opening: ensure no other app is using the camera; try different index (0, 1, 2).
Slow performance: use yolov8n.pt, reduce resolution, or close other heavy apps.
No detections: lower confidence threshold, improve lighting, ensure objects fill the frame.
Model download fails: verify internet connection or manually place yolov8n.pt in the project root.
Import errors: run pip install -r requirements.txt.

🧭 Project Structure (example)

Vison_Learning/
├── streamlit_app.py        # Streamlit web app (main)
├── app.py                  # OpenCV desktop app
├── requirements.txt        # Dependencies
├── README.md               # Project documentation
├── yolov8n.pt              # Model weights (auto-download)
└── screenshots/            # Saved screenshots

📈 Ideas for Improvements / Contributing

Contributions welcome! Suggested enhancements:

Multi-language support (translate detected terms)
Quiz mode (user names the object before reveal)
Analytics dashboard (progress across sessions)
Text-to-speech for pronunciations
Custom model training for domain-specific objects
Mobile/browser optimizations

If you’d like to contribute, fork the repo, open a branch, and submit a PR. Include tests and update the README where appropriate.

📄 License

This project is released under the MIT License. See the LICENSE file for details.