Unofficial re-implementation for MemSeg: A semi-supervised method for image surface defect detection using differences and commonalities

Environments

• Docker image: nvcr.io/nvidia/pytorch:20.12-py3

einops==0.5.0
timm==0.5.4
wandb==0.12.17
omegaconf
imgaug==0.4.0

Process

1. Anomaly Simulation Strategy

• notebook
• Describable Textures Dataset(DTD) [ download ]

2. Model Process

• notebook

Run

Example

python main.py configs=configs.yaml DATASET.target=bottle

Demo

voila "[demo] model inference.ipynb" --port ${port} --Voila.ip ${ip}

Results

• Backbone: ResNet18

target	AUROC-image	AUROC-pixel	AUPRO-pixel
leather	100	98.83	99.09
pill	97.05	98.29	97.96
carpet	99.12	97.54	97.02
hazelnut	100	97.78	99
tile	99.86	99.38	98.81
cable	92.5	82.3	87.31
toothbrush	100	99.28	98.56
transistor	96.5	76.29	86.06
zipper	99.95	97.94	97.26
metal_nut	99.46	88.48	95
grid	99.83	98.37	98.53
bottle	100	98.79	98.36
capsule	95.41	98.43	97.73
screw	94.86	95.08	94
wood	100	97.54	97.62
Average	98.3	94.96	96.15

Citation

@article{DBLP:journals/corr/abs-2205-00908,
  author    = {Minghui Yang and
               Peng Wu and
               Jing Liu and
               Hui Feng},
  title     = {MemSeg: {A} semi-supervised method for image surface defect detection
               using differences and commonalities},
  journal   = {CoRR},
  volume    = {abs/2205.00908},
  year      = {2022},
  url       = {https://doi.org/10.48550/arXiv.2205.00908},
  doi       = {10.48550/arXiv.2205.00908},
  eprinttype = {arXiv},
  eprint    = {2205.00908},
  timestamp = {Tue, 03 May 2022 15:52:06 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2205-00908.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}