|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- anomaly-detection |
|
|
- efficientad |
|
|
- mvtec-ad |
|
|
- cable |
|
|
--- |
|
|
|
|
|
# EfficientAD - Cable |
|
|
|
|
|
EfficientAD model for detecting bent wires, cable swaps, and cut insulation in cables |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Architecture**: EfficientAD (Teacher-Student-Autoencoder) |
|
|
- **Model Size**: Medium (512-dimensional features) |
|
|
- **Dataset**: MVTec AD - Cable |
|
|
- **AU-ROC**: 94.2% |
|
|
- **Training**: Custom training on Apple Silicon (MPS) |
|
|
|
|
|
## Files |
|
|
|
|
|
- `teacher.pth`: Pre-trained teacher network (31MB) |
|
|
- `student.pth`: Trained student network (44MB) |
|
|
- `autoencoder.pth`: Trained autoencoder (4.2MB) |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
import torch |
|
|
|
|
|
# Load models |
|
|
teacher = torch.load('teacher.pth') |
|
|
student = torch.load('student.pth') |
|
|
autoencoder = torch.load('autoencoder.pth') |
|
|
``` |
|
|
|
|
|
## Citation |
|
|
|
|
|
```bibtex |
|
|
@article{efficientad2023, |
|
|
title={EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies}, |
|
|
author={Batzner, Kilian and Heckler, Lars and König, Rebecca}, |
|
|
journal={arXiv preprint arXiv:2303.14535}, |
|
|
year={2023} |
|
|
} |
|
|
``` |
|
|
|
|
|
Generated with Lumina Tech Platform |
|
|
|
