The model supports both mobile inference (ONNX) and web/server inference (PyTorch) and is designed to work in real-world conditions with multiple warning lights visible at once.

🔍 What it does

Detects multiple warning lights in a single dashboard image
Classifies 68 different warning light types
Provides confidence scores
Designed for AI-assisted vehicle diagnostics
Works with:
- 📱 Mobile apps (ONNX)
- 🌐 Web / backend servers (PyTorch)

🖼️ Demo / Screenshots

🧠 Model Architecture

Backbone: ResNet50
Input size: 3 × 224 × 224
Head:
- Dropout (0.3)
- Linear (2048 → 512)
- ReLU
- BatchNorm
- Dropout (0.15)
- Linear (512 → 68)
Parameters: ~24.6M

Architecture definition:
📄 model_architecture.py

📦 Files

File	Description
`car_warning_lights_resnet50.pth`	PyTorch weights (web/server inference)
`car_warning_lights_resnet50.onnx`	ONNX model (mobile / edge inference)
`model_architecture.py`	Model definition
`README.md`	Model documentation

🚀 Usage

PyTorch

python

import torch
from model_architecture import create_model

model = create_model(num_classes=68)
model.load_state_dict(
    torch.load("car_warning_lights_resnet50.pth", map_location="cpu")
)
model.eval()