README.md

metadata

title: Bus Inspection Classifier
emoji: 🚌
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.2.0
app_file: app.py
pinned: false
license: mit

🚌 Bus Inspection Classifier - SigLIP v2 ONNX

An automated bus component classification system using the SigLIP v2 vision model, optimized with ONNX Runtime for fast inference.

🎯 Model Details

• Model: SigLIP v2 (Sigmoid Loss for Language-Image Pre-training)
• Format: ONNX (372.6 MB)
• Input: RGB images (224x224)
• Classes: 18 bus component categories
• Inference Time: ~50-100ms per image (CPU)

📋 Supported Classes

The model can classify the following bus components:

1. AC Mat
2. Alco brake camera
3. Alco-brake device
4. Back windshield
5. Bus back side
6. Bus front side
7. Bus side
8. Cabin
9. Driver grooming
10. First aid kit
11. Floormats & POS
12. Front windshield
13. Hat rack
14. ITMS Device
15. Jack & Spare tyre
16. Luggage compartment
17. RFID Card
18. Seats

🔌 API Usage

Using the Gradio Client (Recommended)

from gradio_client import Client

# Connect to the Space
client = Client("YOUR-USERNAME/bus-inspection")

# Make a prediction
result = client.predict("path/to/bus_image.jpg")

print(result)
# Output: {
#   "class_name": "Bus front side",
#   "confidence": "98.45%",
#   "inference_time_ms": "87.32"
# }

Using Python Requests

import requests
import json
from PIL import Image
import io

# Your Space URL
API_URL = "https://YOUR-USERNAME-bus-inspection.hf.space/api/predict"

# Load image
with open("bus_image.jpg", "rb") as f:
    files = {"data": f}
    response = requests.post(API_URL, files=files)

result = response.json()
print(f"Class: {result['class_name']}")
print(f"Confidence: {result['confidence']}")
print(f"Inference Time: {result['inference_time_ms']} ms")

Using cURL

curl -X POST https://YOUR-USERNAME-bus-inspection.hf.space/api/predict \
  -F "data=@bus_image.jpg"

📊 Response Format

The API returns a JSON object with the following fields:

{
  "class_name": "Bus front side",
  "confidence": "98.45%",
  "inference_time_ms": "87.32"
}

• class_name: The predicted bus component category
• confidence: Model confidence score as a percentage
• inference_time_ms: Total inference time in milliseconds

🚀 Example Use Cases

• Quality Control: Automated inspection of bus components during manufacturing
• Maintenance Checks: Verify proper installation of safety equipment
• Documentation: Automatically categorize bus inspection photos
• Training: Help inspectors identify components correctly

🛠️ Technical Implementation

• Framework: Gradio for UI and API
• Runtime: ONNX Runtime (CPU optimized)
• Preprocessing: SigLIP normalization (mean=0.5, std=0.5)
• Deployment: Hugging Face Spaces with Git LFS

📝 License

MIT License - Free for commercial and personal use

🙏 Acknowledgments

Built with SigLIP (Sigmoid Loss for Language-Image Pre-training) from Google Research.