PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation

Reiss, Tal; Cohen, Niv; Bergman, Liron; Hoshen, Yedid

doi:10.48550/arxiv.2010.05903

Cited by 6 publications

(38 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Analogous pre-training for OCC has been proposed by [3], where they jointly train anomaly detection with the original task, which achieves only limited adaptation success. PANDA [1] proposed techniques based on early stopping and EWC [22], a continual learning method, to mitigate catastrophic collapse. Although PANDA achieved state-of-the-art performance on most datasets, it has yet to solve the problem of catastrophic collapse.…”

Section: Related Workmentioning

confidence: 99%

“…The feature extractor φ is initialized with some pre-trained feature extractor φ 0 (an ImageNet pretrained ResNet was shown by [1] to be very effective). The center can be set to be the mean of the training set pre-trained feature representation:…”

Section: Center Lossmentioning

confidence: 99%

“…However, the proposed architectures were too constrained. PANDA [1] showed that the DeepSVDD feature adaptation does not perform better than linear post-processing of the initial feature extractor φ 0 . Instead, they use the center loss without any architectural modifications, and propose techniques based on early stopping and continual learning to mitigate catastrophic collapse.…”

Section: Center Lossmentioning

confidence: 99%

“…1), which learns representations by minimizing the Euclidean distance between normal representations and the center. PANDA [1] has shown that the optimized center loss results in high anomaly detection performance. Furthermore, forcing the representations of the normal training data to be uniformly distributed on the unit sphere is not well suited for OCC, since every anomalous sample x / ∈ X train will have a nearby normal sample x ∈ X train .…”

Section: Mean-shifted Loss: Modifying the Contrastive Loss For Occ Tr...mentioning

confidence: 99%

See 3 more Smart Citations

Mean-Shifted Contrastive Loss for Anomaly Detection

Reiss¹,

Hoshen²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Deep anomaly detection methods learn representations that separate between normal and anomalous samples. Very effective representations are obtained when powerful externally trained feature extractors (e.g. ResNets pre-trained on Ima-geNet) are fine-tuned on the training data which consists of normal samples and no anomalies. However, this is a difficult task that can suffer from catastrophic collapse, i.e. it is prone to learning trivial and non-specific features. In this paper, we propose a new loss function which can overcome failure modes of both center-loss and contrastive-loss methods. Furthermore, we combine it with a confidence-invariant angular center loss, which replaces the Euclidean distance used in previous work, that was sensitive to prediction confidence. Our improvements yield a new anomaly detection approach, based on Mean-Shifted Contrastive Loss, which is both more accurate and less sensitive to catastrophic collapse than previous methods. Our method achieves state-of-the-art anomaly detection performance on multiple benchmarks including 97.5% ROC-AUC on the CIFAR-10 dataset 1 . IntroductionAnomaly detection is a fundamental task for intelligent agents that aims to detect if an observed pattern is normal or anomalous (unusual or unlikely). Anomaly detection has broad applications in scientific and industrial tasks such as detecting new physical phenomena (black holes, supernovae) or genetic mutations, as well as production line inspection and video surveillance. Due to the significance of the task, many efforts have been focused on automatic anomaly detection, particularly on statistical and machine learning methods. A common paradigm used by many anomaly detection methods is measuring the probability of samples and assigning high-probability samples as normal and low-probability samples as anomalous. The quality of the density estimators is closely related to the quality of features used to represent the data. Classical methods used statistical estimators such as K nearest-neighbors (kNN) or Gaussian mixture models (GMMs) on raw features, however this often results in sub-optimal results on high-dimensional data such as images. Many recent methods, learn features in a self-supervised way and use them in order to detect anomalies. Their main weakness is that anomaly detection datasets are typically small and do not include anomalous samples resulting in weak features. An alternative direction, which achieved better results, is to transfer features learned from auxiliary tasks on large-scale external datasets such as ImageNet classification. It was found that fine-tuning the pre-trained features on the normal training data can result in significant performance improvements, however it is quite challenging. The main issue with fine-tuning on one-class classification (OCC) tasks such as anomaly detection is catastrophic collapse i.e. after an initial improvement in efficacy, the features degrade and become uninformative. This phenomenon is caused by trivial solutions allowed by OCC tasks such as t...

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Center Lossmentioning

confidence: 99%

Section: Center Lossmentioning

confidence: 99%

Section: Mean-shifted Loss: Modifying the Contrastive Loss For Occ Tr...mentioning

confidence: 99%

See 2 more Smart Citations

Mean-Shifted Contrastive Loss for Anomaly Detection

Reiss¹,

Hoshen²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Leveraging Pre-trained CNN: Usually the sample scale of anomaly detection datasets is relatively smaller than Im-ageNet dataset [17], so it is inadequate to learning a good representation from scratch. Many methods have also been proposed for anomaly detection [18,19,20,21] or segmentation [2,22,23,24,25] by using deep representations pretrained on ImageNet [17]. They compare patch features at the same position from target and normal image to obtain pixellevel anomaly map.…”

Section: Introductionmentioning

confidence: 99%

Anomaly Detection via Self-organizing Map

Li¹,

Jiang²,

et al. 2021

Preprint

View full text Add to dashboard Cite

Anomaly detection plays a key role in industrial manufacturing for product quality control. Traditional methods for anomaly detection are rule-based with limited generalization ability. Recent methods based on supervised deep learning are more powerful but require large-scale annotated datasets for training. In practice, abnormal products are rare thus it is very difficult to train a deep model in a fully supervised way. In this paper, we propose a novel unsupervised anomaly detection approach based on Self-organizing Map (SOM). Our method, Self-organizing Map for Anomaly Detection (SO-MAD) maintains normal characteristics by using topological memory based on multi-scale features. SOMAD achieves state-of-the-art performance on unsupervised anomaly detection and localization on the MVTec dataset.

show abstract

Anomaly Detection and Biomarkers Localization in Retinal OCT Scans

Tiosano,

Abutbul,

Lender

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Anomaly detection combined with localization in retinal scans can help identify retinal anomalies scans and localize pathologies that might otherwise be difficult to detect. We designed a novel approach for detecting anomalies and localization by applying AI-based tools to optical coherence tomography (OCT) scans in retinal disease. High-resolution OCT-scans from the public and a local dataset were used in four state-of-the-art self-supervised frameworks. The backbone for these frameworks was a pre-trained convolutional neural network, which allowed us to extract meaningful features from the OCT-images. Anomalous images included choroidal neovascularization, diabetic macular edema, and drusen. The resulting anomaly detectors were then evaluated using area under the receiver operating characteristic curve (ROC AUC) scores, F1 scores, and accuracy. Approximately 30000 OCT-images were used. The best-performing anomaly detectors had an ROC AUC score of 0.99, and all frameworks achieved high performance and generalized well to various retinal diseases. Using pre-trained feature extractors, the frameworks tested here can be generalized to retinal OCT-scans, yielding high image-level ROC AUC scores. The localization results obtained using these frameworks can successfully capture areas indicating the presence of retinal pathology; moreover, these frameworks may also reveal new biomarkers. Finally, these frameworks can be integrated into clinical decision-making and automated screening systems, thereby facilitating treatment management.

show abstract

PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation

Cited by 6 publications

References 16 publications

Mean-Shifted Contrastive Loss for Anomaly Detection

Mean-Shifted Contrastive Loss for Anomaly Detection

Anomaly Detection via Self-organizing Map

Anomaly Detection and Biomarkers Localization in Retinal OCT Scans

Contact Info

Product

Resources

About