Awesome

This project is an unofficial implementation of "EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies".

Datasets

./data

• ImageNet

  • n01440764
  • n01443537 ...

• MVTec_AD

  • bottle
    • ground_truth
    • test
    • train
  • cable
    • ground_truth
    • test
    • train ...

• result

Quick start

1. Install PyTorch environment

conda activate <your_env>
pip install -r requirements.txt

1. Distill a PDN architecture teacher network from wide_resnet101

python distillaion_training.py

2. train the student network and autoencoder network

python train_reduced_student.py  -c configs/mvtec_train.yaml

Pretrain Weights

Download pretrain weights from release page.

Some results

MVTec bottle