Live Computer Vision Project (ResNet-50 CIFAR-100)

A computer vision application powered by a fine-tuned ResNet-50 deep learning model trained on CIFAR-100. Supports real-time webcam inference, desktop OpenCV window with dynamic HUD, image snapshot saving, and an interactive FastAPI Web Computer Vision Studio.

🚀 Features

Trained ResNet-50 Model Integration: Loads resnet50_cifar100_finetuned.pth with hardware acceleration auto-detection (Apple Silicon Metal MPS, NVIDIA CUDA, or CPU).
Desktop OpenCV Feed (--mode desktop):
- Live webcam stream with real-time target region box.
- Interactive Heads-Up Display (HUD) with Top-5 probability horizontal bar charts.
- Keybindings: c (predict frame), r (toggle continuous real-time mode), s (save snapshot), q (quit).
Web Computer Vision Studio (--mode web):
- Web UI powered by FastAPI and Uvicorn.
- Real-time webcam classification directly in browser.
- Drag-and-Drop Image Uploader for testing static files.
- Glassmorphic dark theme UI with animated confidence indicators.
Single Image Predictor (predict_image.py):
- CLI command to classify any image file.

📂 Project Files

main.py: Unified CLI entry point for Desktop and Web modes.
desktop_cv.py: Desktop OpenCV webcam application with HUD overlays.
web_app.py: FastAPI web server and single-page Web Vision Studio.
model_loader.py: Model loading, preprocessing pipeline, and inference engine.
cifar100_labels.py: Official 100 CIFAR-100 class names and supercategory mapping.
predict_image.py: CLI tool for single image classification.
resnet50_cifar100_finetuned.pth: Fine-tuned PyTorch ResNet-50 weights.

🛠️ Requirements & Setup

Make sure PyTorch, torchvision, OpenCV, and FastAPI are installed:

pip install torch torchvision opencv-python pillow fastapi uvicorn

🎮 How to Run

1. Run Desktop Webcam Mode (OpenCV Window)

python3 main.py --mode desktop

Press c: Capture & predict current frame
Press r: Toggle continuous real-time inference
Press s: Save snapshot with HUD overlay to snapshots/ folder
Press q: Quit application

2. Run Web Studio Server (Browser Mode)

python3 main.py --mode web --port 8000

Open your browser at http://127.0.0.1:8000 to access the interactive web interface.

3. Predict Single Image File

python3 predict_image.py /path/to/your/image.jpg

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support