Unofficial PyTorch implementation of Towards Total Online Unsupervised Anomaly Detection and Localization in Industrial Vision (LeMO). This work is part of a thesis in Artificial Intelligence.
Install packages with:
$ pip install -r requirements.txt
Prepare industrial image as:
train data:
dataset_path/class_name/train/good/any_filename.png
[...]
test data:
dataset_path/class_name/test/good/any_filename.png
[...]
dataset_path/class_name/test/defect_type/any_filename.png
[...]
python trainer_lemo.py --class_name all --data_path [/path/to/dataset/] --results_path [/path/to/results/] --cnn wrn50_2 --size 224 --gamma_c 1 --gamma_d 1 --loss NCENEW --memory_update kmeans
Official LeMO | Ours | |
---|---|---|
Image AUROC | 0.972 | 0.956 |
Pixel AUROC | 0.976 | 0.970 |
For more details about the model performances see the LeMO_report.pdf, chapter 5.2.5.