Industrial KNN-based Anomaly Detection
$ streamlit run streamlit_app.py
This repo aims to reproduce the results of the following KNN-based anomaly detection methods:
- SPADE (Cohen et al. 2021) - knn in z-space and distance to feature maps
- PaDiM* (Defard et al. 2020) - distance to multivariate Gaussian of feature maps
- PatchCore (Roth et al. 2021) - knn distance to avgpooled feature maps
* actually does not have any knn mechanism, but shares many things implementation-wise.
Install
$ pipenv install -r requirements.txt
Note: I used torch cu11 wheels.
Usage
CLI:
$ python indad/run.py METHOD [--dataset DATASET]
Results can be found under ./results/
.
Code example:
from indad.model import SPADE
model = SPADE(k=5, backbone_name="resnet18")
# feed healthy dataset
model.fit(...)
# get predictions
img_lvl_anom_score, pxl_lvl_anom_score = model.predict(...)
Custom datasets
ποΈ
Check out one of the downloaded MVTec datasets. Naming of images should correspond among folders. Right now there is no support for no ground truth pixel masks.
πdatasets
β πyour_custom_dataset
β£ π ground_truth/defective
β β£ π defect_type_1
β β π defect_type_2
β£ π test
β β£ π defect_type_1
β β£ π defect_type_2
β β π good
β π train/good
$ python indad/run.py METHOD --dataset your_custom_dataset
Results
Image-level
class | SPADE |
SPADE |
PaDiM |
PaDiM |
PatchCore |
PatchCore |
---|---|---|---|---|---|---|
bottle | - | 98.3 | 98.3 | 99.9 | 100.0 | 100.0 |
cable | - | 88.1 | 96.7 | 87.8 | 99.5 | 96.2 |
capsule | - | 80.4 | 98.5 | 87.6 | 98.1 | 95.3 |
carpet | - | 62.5 | 99.1 | 99.5 | 98.7 | 98.7 |
grid | - | 25.6 | 97.3 | 95.5 | 98.2 | 93.0 |
hazelnut | - | 92.8 | 98.2 | 86.1 | 100.0 | 100.0 |
leather | - | 85.6 | 99.2 | 100.0 | 100.0 | 100.0 |
metal_nut | - | 78.6 | 97.2 | 97.6 | 100.0 | 98.3 |
pill | - | 78.8 | 95.7 | 92.7 | 96.6 | 92.8 |
screw | - | 66.1 | 98.5 | 79.6 | 98.1 | 96.7 |
tile | - | 96.4 | 94.1 | 99.5 | 98.7 | 99.0 |
toothbrush | - | 83.9 | 98.8 | 94.7 | 100.0 | 98.1 |
transistor | - | 89.4 | 97.5 | 95.0 | 100.0 | 99.7 |
wood | - | 85.3 | 94.7 | 99.4 | 99.2 | 98.8 |
zipper | - | 97.1 | 98.5 | 93.8 | 99.4 | 98.4 |
averages | 85.5 | 80.6 | 97.5 | 93.9 | 99.1 | 97.7 |
Pixel-level
class | SPADE |
SPADE |
PaDiM |
PaDiM |
PatchCore |
PatchCore |
---|---|---|---|---|---|---|
bottle | 97.5 | 97.7 | 94.8 | 97.6 | 98.6 | 97.8 |
cable | 93.7 | 94.4 | 88.8 | 95.5 | 98.5 | 97.4 |
capsule | 97.6 | 98.7 | 93.5 | 98.1 | 98.9 | 98.3 |
carpet | 87.4 | 99.0 | 96.2 | 98.7 | 99.1 | 98.3 |
grid | 88.5 | 96.4 | 94.6 | 96.4 | 98.7 | 96.7 |
hazelnut | 98.4 | 98.4 | 92.6 | 97.3 | 98.7 | 98.1 |
leather | 97.2 | 99.1 | 97.8 | 98.6 | 99.3 | 98.4 |
metal_nut | 99.0 | 96.1 | 85.6 | 95.8 | 98.4 | 96.2 |
pill | 99.1 | 93.5 | 92.7 | 94.4 | 97.6 | 98.7 |
screw | 98.1 | 98.9 | 94.4 | 97.5 | 99.4 | 98.4 |
tile | 96.5 | 93.1 | 86.0 | 92.6 | 95.9 | 94.0 |
toothbrush | 98.9 | 98.9 | 93.1 | 98.5 | 98.7 | 98.1 |
transistor | 97.9 | 95.8 | 84.5 | 96.9 | 96.4 | 97.5 |
wood | 94.1 | 94.5 | 91.1 | 92.9 | 95.1 | 91.9 |
zipper | 96.5 | 98.3 | 95.9 | 97.0 | 98.9 | 97.6 |
averages | 96.9 | 96.6 | 92.1 | 96.5 | 98.1 | 97.2 |
PatchCore-10 was used.
Hyperparams
The following parameters were used to calculate the results. They more or less correspond to the parameters used in the papers.
spade:
backbone: wide_resnet50_2
k: 50
padim:
backbone: wide_resnet50_2
d_reduced: 250
epsilon: 0.04
patchcore:
backbone: wide_resnet50_2
f_coreset: 0.1
n_reweight: 3
Progress
- Datasets
- Code skeleton
- Config files
- CLI
- Logging
- SPADE
- PADIM
- PatchCore
- Add custom dataset option
- Add dataset progress bar
- Add schematics
- Unit tests
Design considerations
- Data is processed in single images to avoid batch statistics interference.
- I decided to implement greedy kcenter from scratch and there is room for improvement.
torch.nn.AdaptiveAvgPool2d
for feature map resizing,torch.nn.functional.interpolate
for score map resizing.- GPU is used for backbones and coreset selection. GPU coreset selection currently runs at:
- 400-500 it/s @ float32 (RTX3080)
- 1000+ it/s @ float16 (RTX3080)
Acknowledgements
- hcw-00 for tipping
sklearn.random_projection.SparseRandomProjection
References
SPADE:
@misc{cohen2021subimage,
title={Sub-Image Anomaly Detection with Deep Pyramid Correspondences},
author={Niv Cohen and Yedid Hoshen},
year={2021},
eprint={2005.02357},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
PaDiM:
@misc{defard2020padim,
title={PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization},
author={Thomas Defard and Aleksandr Setkov and Angelique Loesch and Romaric Audigier},
year={2020},
eprint={2011.08785},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
PatchCore:
@misc{roth2021total,
title={Towards Total Recall in Industrial Anomaly Detection},
author={Karsten Roth and Latha Pemula and Joaquin Zepeda and Bernhard SchΓΆlkopf and Thomas Brox and Peter Gehler},
year={2021},
eprint={2106.08265},
archivePrefix={arXiv},
primaryClass={cs.CV}
}