car-damage-classifier

🚗 Car Damage Classifier & Cost Estimator (EfficientNetV2-M)

A production-grade EfficientNetV2-M computer vision model fine-tuned for automated vehicle damage classification and repair cost range estimation for automotive insurance claim pipelines.

Identifies 7 damage severity categories and maps each prediction to estimated repair costs in Polish Złoty (PLN).

---

🎯 Model Capabilities

• Damage Classification: 7 distinct damage categories including structural, component, and catastrophe damage.
• Cost Estimation: Integrates historical insurer claims cost intervals (PLN) for fast claims pre-assessment.
• Dual Formats:
  • efficientnet_v2_m_best.pth — PyTorch weights.
  • efficientnet_v2_m_damage.onnx + .onnx.data — High-performance ONNX model for TensorRT / C++ / Edge inference.
  • damage_metadata.json — Class names and corresponding repair cost intervals.

---

📊 Damage Classes & Estimated Repair Cost Ranges

---

📈 Evaluation & Benchmark Metrics

• Overall Test Accuracy: 100% (across 420 balanced evaluation samples).
• Macro F1 Score: 1.000
• Weighted F1 Score: 1.000
• Business Misclassification Cost: 0.00 PLN on test set (zero critical under-estimation of severe/total loss claims).

---

🚀 Quick Start (Inference)

1. Using ONNX Runtime (No PyTorch needed)

pip install onnxruntime pillow numpy huggingface_hub

import json
import numpy as np
from PIL import Image
import onnxruntime as ort
from huggingface_hub import hf_hub_download

# 1. Download model & metadata
model_path = hf_hub_download(repo_id="BiernyVR/car-damage-classifier", filename="efficientnet_v2_m_damage.onnx")
data_path = hf_hub_download(repo_id="BiernyVR/car-damage-classifier", filename="efficientnet_v2_m_damage.onnx.data")
meta_path = hf_hub_download(repo_id="BiernyVR/car-damage-classifier", filename="damage_metadata.json")

with open(meta_path, "r", encoding="utf-8") as f:
    meta = json.load(f)

classes = meta["classes"]
cost_ranges = meta["cost_ranges_pln"]

# 2. Preprocess image
img = Image.open("damaged_car.jpg").convert("RGB").resize((224, 224), Image.Resampling.BILINEAR)
arr = (np.array(img, dtype=np.float32) / 255.0 - [0.485, 0.456, 0.406]) / [0.229, 0.224, 0.225]
tensor = np.expand_dims(np.transpose(arr, (2, 0, 1)), axis=0).astype(np.float32)

# 3. Run Inference
session = ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])
logits = session.run(None, {"input": tensor})[0][0]
probs = np.exp(logits - np.max(logits))
probs /= probs.sum()

pred_idx = np.argmax(probs)
pred_class = classes[pred_idx]
cost = cost_ranges[pred_class]
print(f"Diagnosis: {pred_class} ({probs[pred_idx]*100:.1f}%)")
print(f"Estimated Cost: {cost[0]} - {cost[1]} PLN")

2. Standalone CLI

python infer.py --image sample_dent.jpg --topk 3

---

📦 Repository Contents

• efficientnet_v2_m_best.pth: PyTorch weights checkpoint.
• efficientnet_v2_m_damage.onnx + .onnx.data: Exported ONNX model.
• damage_metadata.json: Classes and PLN cost boundaries.
• confusion_matrix.png, business_cost_matrix.png, per_class_metrics.png, training_curves.png: Evaluation and training plots.
• sample_dent.jpg, sample_severe.jpg: Example test images.
• infer.py: Standalone CLI testing script.