Towards Visually Explaining Variational Autoencoders

Liu, Wenqian; Li, Runze; Zheng, Meng; Karanam, Srikrishna; Wu, Ziyan; Bhanu, Bir; Radke, Richard J.; Camps, Octavia

doi:10.1109/cvpr42600.2020.00867

Cited by 160 publications

(110 citation statements)

References 33 publications

Supporting

Mentioning

108

Contrasting

Order By: Relevance

“…Consequently, previous approaches have attempted to enhance the models for a better reconstruction performance. For example, well-known models were combined [1], cross-channel interactions were enabled [3], a deep convolutional generative adversarial network was used [4], a de-noising technique was proposed [5], pixel-level method and patch-level method were mixed [8], an attention technique was utilized [10], [13], and encoder-decoderencoder sub-networks were employed [15]. In [9], the reconstruction performance is improved by iteratively updating the input image using a gradient descent.…”

Section: Related Work a Unsupervised Anomaly Localizationmentioning

confidence: 99%

See 1 more Smart Citation

Patch-Level Operation With Adaptive Patch Control for Improving Anomaly Localization

et al. 2021

View full text Add to dashboard Cite

In realizing unsupervised pixel-precise anomaly localization by utilizing a generative model, a reference image must be generated (for comparison with an input image) by transforming abnormal patterns of an input image, if any, into normal patterns. In this study, a patch-level operation with adaptive patch control is proposed to improve anomaly localization by generating a better reference image. As a way to exploit a generative model, we divide an image into non-overlapped patches of the same size, generate patch-level reference images, and stitch the patch-level reference images into a single reference image. We then conduct anomaly localization by comparing an input image with the stitched, reconstructed image. To effectively apply the patch-level operation, we propose adaptive patch control to determine the number of non-overlapped patches to be applied. For this, we synthesize defective images using normal images and examine how well the candidate methods with different numbers of patches remove the synthesized defects. In the same way, we utilize adaptive patch control to select a promising model among the candidate generative models. Based on experiments conducted using the MVTec Anomaly Detection dataset, we demonstrate that our method outperforms previous existing methods. Under a real-world scenario, our method shows ROC AUC of 0.926, in contrast to the best value of 0.893 from existing studies. Furthermore, we prove the feasibility of the adaptive patch control by showing that the removal of the synthesized defects and the anomaly localization for real defective images are highly correlated.

show abstract

Section: Related Work a Unsupervised Anomaly Localizationmentioning

confidence: 99%

“…This method aims to generate a reference image by transforming abnormal patterns of a test image, if any, into normal patterns observed in normal training images. Efforts have been made to develop effective models and techniques for achieving better reference images [1], [3], [5], [10], [13].…”

Section: Introductionmentioning

confidence: 99%

Patch-Level Operation With Adaptive Patch Control for Improving Anomaly Localization

et al. 2021

View full text Add to dashboard Cite

show abstract

“…This method classifies an abnormal sample using the difference in a loss function or mutual information between normal sample and abnormal sample. The autoencoder-based method is an unsupervised-based method composed of an encoder and a decoder [15] - [22]. The autoencoder-based anomaly detection method uses the image difference between input and output images [16], [18], [20], a latent space-based score [15], [17], a loss-based score [19], and the method using the autoencoder-based GAN structure [20].…”

Section: Introductionmentioning

confidence: 99%

Anomaly Detection Using Autoencoder With Feature Vector Frequency Map

Kim

Park

2021

IEEE Access

View full text Add to dashboard Cite

Anomaly detection uses various machine learning techniques to identify and classify defective data on the production line. The autoencoder-based anomaly detection method is an unsupervised method that classifies abnormal samples using an autoencoder trained only from normal samples and is useful in environments where it is difficult to obtain abnormal samples. This method uses an abnormal score based on the reconstruction loss function, making it difficult to detect defects, such as stains, having a similar texture to a normal sample. To solve this problem, we propose an anomaly detection method using a vector quantized variational autoencoder and a feature vector frequency map. We use the prototype vector histogram and its frequency for anomaly detection instead of the reconstruction loss function. The prototype vector histogram is obtained from the vector quantized variational autoencoder's codebook in the training stage. The feature vector frequency map of the input image is generated using the prototype vector histogram in the inference stage. We calculated the abnormal score using the generated frequency map and classified the abnormal samples. The experimental results showed that the proposed method has a higher Area Under Receiver Operating Characteristics (AUROC) than the previous method in stain and scratch defects.

show abstract

“…We achieve state-of-the-art textural anomaly segmentation results using a relatively simple architecture. In contrast, other methods often employ more complex architectures such as the Generative Adversarial Network (GAN) [6], [7], [13], or the Variational Autoencoder (VAE) [8], [10]. In addition, our proposed method is computationally light-weight and stable during training.…”

Section: Introductionmentioning

confidence: 99%

“…Masci et al [15] adapt this architecture by using convolutional layers. These convolutional autoencoders are better suited to operating on image data and are often applied to anomaly detection in images and videos [5], [8], [10], [16].…”

Section: Introductionmentioning

confidence: 99%

Autoencoders Without Reconstruction for Textural Anomaly Detection

Adey

Akçay

Bordewich

et al. 2021

2021 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

Automatic anomaly detection in natural textures is a key component within quality control for a range of high-speed, high-yield manufacturing industries that rely on camera-based visual inspection techniques. Targeting anomaly detection through the use of autoencoder reconstruction error readily facilitates training on an often more plentiful set of non-anomalous samples, without the explicit need for a representative set of anomalous training samples that may be difficult to source. Unfortunately, autoencoders struggle to reconstruct high-frequency visual information and therefore, such approaches often fail to achieve a low enough reconstruction error for non-anomalous pixels. In this paper, we propose a new approach in which the autoencoder is trained to directly output the desired per-pixel measure of abnormality without first having to perform reconstruction. This is achieved by corrupting training samples with noise and then predicting how pixels need to be shifted so as to remove the noise. Our direct approach enables the model to compress anomaly scores for normal pixels into a tight bound close to zero, resulting in very clean anomaly segmentations that significantly improve performance. We also introduce the Reflected ReLU output activation function that better facilitates training under this direct regime by leaving values that fall within the image dynamic range unmodified. Overall, an average area under the ROC curve of 96% is achieved on the texture classes of the MVTecAD benchmark dataset, surpassing that achieved by all current state-of-the-art methods.

show abstract

Towards Visually Explaining Variational Autoencoders

Cited by 160 publications

References 33 publications

Patch-Level Operation With Adaptive Patch Control for Improving Anomaly Localization

Patch-Level Operation With Adaptive Patch Control for Improving Anomaly Localization

Anomaly Detection Using Autoencoder With Feature Vector Frequency Map

Autoencoders Without Reconstruction for Textural Anomaly Detection

Contact Info

Product

Resources

About