representation-based approaches extract normal image features
with a deep convolutional neural network and characterize
the corresponding distribution through non-parametric
distribution estimation methods.
non-parametric distribution estimation method로 분포를 추정함⇒kNN,Mahalanobis distance
좀 더 flexible한 장점은 있지만 연산량 증가 및 모델에 대한 속성이 implict하여 명확한 설명을 하기 쉽지 않음
The anomaly score is calculated
by measuring the distance between the feature of the
test image and the estimated distribution. However, current
methods can not effectively map image features to a tractable
base distribution and ignore the relationship between local
and global features which are important to identify anomalies.
PatchCore와 같이 feature matching 기반 모델들은 다루기 쉬운분포(정규분포)등에 feature를 matching하지 못한다.(위의 non-parametric model의 문제점을 이렇게 설명하였다.)
또한 local feature와 global feature간의 relationship를 고려하지 않는다.
we propose FastFlow implemented with 2D
normalizing flows and use it as the probability distribution estimator.
Our FastFlow can be used as a plug-in module with
arbitrary deep feature extractors such as ResNet and vision
transformer for unsupervised anomaly detection and localization.
In training phase, FastFlow learns to transform the input
visual feature into a tractable distribution and obtains the
likelihood to recognize anomalies in inference phase.
