valiente/gate-classifier-onnx-intel repository overview

⁠VisionGate - Docker Deployment

A MobileNetV3-based gate classification service containerized with Docker and GPU support.

⁠🚀 Quick Start

⁠Prerequisites

• Docker installed
• NVIDIA Docker runtime (for GPU support)
• Trained model file (gate_mobilenetv3.pth)

⁠Build and Run

Option 1: Using Docker Compose (Recommended)

docker-compose up --build

Option 2: Using Build Scripts

# Linux/Mac
chmod +x deploy.sh
./deploy.sh

# Windows
deploy.bat

Option 3: Manual Docker Commands

# Build
docker build -t gate-classifier .

# Run
docker run -d --name gate-classifier-app --gpus all -p 8000:8000 gate-classifier

⁠🌐 API Endpoints

• Health Check: GET /health
• Classification: POST /classify (upload image)
• API Info: GET /
• Interactive Docs: http://localhost:8000/docs

⁠📋 Usage Example

import requests

# Test the API
url = "http://localhost:8000/classify"
with open("gate_image.jpg", "rb") as f:
    files = {"file": f}
    response = requests.post(url, files=files)
    print(response.json())

⁠🔧 Configuration

⁠Environment Variables

• CUDA_VISIBLE_DEVICES: GPU device selection (default: 0)

⁠Model Files Required

• gate_mobilenetv3.pth (required)
• gate_mobilenetv3_best.pth (optional, preferred if available)

⁠📊 Container Specs

• Base Image: nvidia/cuda:12.1-runtime-ubuntu20.04
• Python: 3.8+
• PyTorch: CUDA 12.1 compatible
• Input Size: 384x384 pixels
• Port: 8000

⁠🛠 Development Commands

# View logs
docker logs -f gate-classifier-app

# Shell access
docker exec -it gate-classifier-app /bin/bash

# Stop container
docker stop gate-classifier-app

# Remove container
docker rm gate-classifier-app

⁠🎯 Performance

• GPU Inference: ~5-10ms per image
• CPU Inference: ~0.5-1s per image
• Model Size: ~170MB
• Memory Usage: ~40MB GPU memory