THE MVTEC ANOMALY DETECTION DATASET (MVTEC AD)

MVTec AD is a dataset for benchmarking anomaly detection methods with a focus on industrial inspection. It contains over 5000 high-resolution images divided into fifteen different object and texture categories. Each category comprises a set of defect-free training images and a test set of images with various kinds of defects as well as images without defects.

conda create -n anomaly-detection python=3.8
conda activate anomaly-detection
pip install -r requirements.txt

For each new dataset, the data consist of three folders:

• train, which contains the (defect-free) training images
• test, which contains the test images
• ground_truth, which contains the pixel-precise annotations of anomalous regions

python train.py --config "configs/patchcore_grid.yaml" --model "patchcore"

or download pretrained models

bash download_pretrained.sh

python script_inference.py --config "configs/patchcore_hazelnut.yaml" --weight "models/patchcore_hazelnut.ckpt" --image "samples/007_hazelnut.png"

or just simple:

python script_inference.py

python demo.py

Open local URL: http://127.0.0.1:7860

Sample:

python app.py

Open local URL: http://127.0.0.1:5000

Default account login:

• Username: aicamp_batch9
• Password: 123456

Homepage:

Train:

Inference:

Database:

docker build -t anomaly:v1 .
docker run anomaly:v1

or just simple

docker-compose up

First: Create EC2 instance

Second: git clone and install related packages

git clone https://github.com/vnk8071/anomaly-detection-in-industry-manufacturing.git

sh download_pretrained.sh

Next: install Miniconda and Docker engine

• Miniconda: Follow link https://varhowto.com/install-miniconda-ubuntu-18-04/
• Docker engine: Follow link https://docs.docker.com/engine/install/ubuntu/

docker-compose up

Final: access link http://user-IPv4-public-ec2-aws