---
tags:
- computer-vision
- defect-detection
- pytorch
- onnx
library_name: pytorch
---

# Defect Classifier (ResNet18)

Steel surface defect classification model trained on NEU-DET dataset.

## Model Details

- **Architecture**: ResNet18
- **Classes**: 6 defect types (crazing, inclusion, patches, pitted_surface, rolled-in_scale, scratches)
- **Input**: 224x224 RGB images
- **Formats**: PyTorch (.pth), ONNX (.onnx)

## Files

- `pytorch_model.pth`: PyTorch checkpoint
- `model.onnx`: ONNX format for deployment
- `metadata.json`: Model configuration
- `training_history.json`: Training metrics

## Usage

### PyTorch

```python
import torch
from torchvision import models
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(repo_id="Seif-melz/defect-classifier-resnet18", filename="pytorch_model.pth")

# Load model
model = models.resnet18()
model.fc = torch.nn.Linear(model.fc.in_features, 6)
checkpoint = torch.load(model_path, map_location='cpu')
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()
```

### ONNX

```python
import onnxruntime as ort
from huggingface_hub import hf_hub_download

# Download ONNX model
model_path = hf_hub_download(repo_id="Seif-melz/defect-classifier-resnet18", filename="model.onnx")

# Load with ONNX Runtime
session = ort.InferenceSession(model_path)
```

## Training

See training history in `training_history.json` for detailed metrics.