Anomaly Detection in Nanofibrous Materials by CNN-Based Self-Similarity

Napoletano, Paolo; Piccoli, Flavio; Schettini, Raimondo

doi:10.3390/s18010209

Cited by 246 publications

(165 citation statements)

References 52 publications

Supporting

Mentioning

154

Contrasting

Unclassified

Order By: Relevance

“…For false positive rates as low as 5%, more than 50% of the defects have an overlap with the ground truth that is larger than 91%. This outperforms the results achieved by Napoletano et al (2018), who report a minimal overlap of 85% in this setting.…”

Section: Resultscontrasting

confidence: 67%

“…Using SSIM as the loss and evaluation metric outperforms all other tested architectures significantly. By merely changing the loss function, the achieved AUC improves from 0.688 to 0.966 on the dataset of nanofibrous materials, which is comparable to the state-of-the-art by Napoletano et al (2018), where values of up to 0.974 are reported. In contrast to this method, autoencoders do not rely on any model priors such as handcrafted features or pretrained networks.…”

Section: Resultsmentioning

confidence: 59%

“…Using the optimal setup of d = 500, k = 11, and patch size 128 × 128, a forward pass through our architecture takes 2.23 ms on a Tesla V100 GPU. Patch-by-patch evaluation of an entire image of the NanoTWICE dataset takes 3.61 s on average, which is significantly faster than the runtimes reported by Napoletano et al (2018). Their approach requires between 15 s and 55 s to process a single input image.…”

Section: Resultsmentioning

confidence: 99%

“…Here, we limit ourselves to an overview of methods that attempt to tackle the latter problem. Napoletano et al (2018) extract features from a CNN that has been pretrained on a classification task. The features are clustered in a dictionary during training and anomalous structures are identified when the extracted features strongly deviate from the learned cluster centers.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders

Bergmann

Löwe

Fauser

et al. 2019

Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Application

300

230

View full text Add to dashboard Cite

Convolutional autoencoders have emerged as popular methods for unsupervised defect segmentation on image data. Most commonly, this task is performed by thresholding a per-pixel reconstruction error based on an p -distance. This procedure, however, leads to large residuals whenever the reconstruction includes slight localization inaccuracies around edges. It also fails to reveal defective regions that have been visually altered when intensity values stay roughly consistent. We show that these problems prevent these approaches from being applied to complex real-world scenarios and that they cannot be easily avoided by employing more elaborate architectures such as variational or feature matching autoencoders. We propose to use a perceptual loss function based on structural similarity that examines inter-dependencies between local image regions, taking into account luminance, contrast, and structural information, instead of simply comparing single pixel values. It achieves significant performance gains on a challenging real-world dataset of nanofibrous materials and a novel dataset of two woven fabrics over state-of-the-art approaches for unsupervised defect segmentation that use per-pixel reconstruction error metrics.

show abstract

Section: Resultscontrasting

confidence: 67%

Section: Resultsmentioning

confidence: 59%

Section: Resultsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders

Bergmann

Löwe

Fauser

et al. 2019

Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Application

300

230

View full text Add to dashboard Cite

show abstract

“…copy-move [1], [21], seam carving [22]. Despite the tremendous progress so far, much potential and many more discoveries lie ahead because of the breakthrough in deep learning [23], [24], many CNN-based methods are investigated and achieve significant improvements [2], [5], [9], [25]. However, as we describe in Section I, these methods all investigate individual images and can not provide the source of splicing images.…”

Section: Related Workmentioning

confidence: 99%

Constrained Image Splicing Detection and Localization With Attention-Aware Encoder-Decoder and Atrous Convolution

Liu

Zhao

2020

IEEE Access

View full text Add to dashboard Cite

Constrained image splicing detection and localization (CISDL) is a newly formulated image forensics task and plays an important role in verifying the generating process of a forged image. CISDL conducts dense matching between two investigated images and detects whether one image has forged regions pasted from the other. In this work, we introduce a novel attention-aware encoder-decoder deep matching network named as AttentionDM for CISDL. An encoder-decoder with atrous convolution is newly designed for hierarchical features dense matching and fine-grained masks generation. A novel attention-aware correlation computation module is built on normalization operations and informative features recalibration with channel attention blocks. Last but not least, VGG and ResNets are respectively formulated as feature extractors for comprehensive comparisons in CISDL. Extensive experiments demonstrate the superior performance of AttentionDM over the state-of-the-art methods.

show abstract

Attention Guided Anomaly Localization in Images

Venkataramanan

Peng

Singh

et al. 2020

Lecture Notes in Computer Science

162

136

View full text Add to dashboard Cite

Anomaly detection and localization is a popular computer vision problem involving detecting anomalous images and localizing anomalies within them. However, this task is challenging due to the small sample size and pixel coverage of the anomaly in real-world scenarios. Prior works need to use anomalous training images to compute a threshold to detect and localize anomalies. To remove this need, we propose Convolutional Adversarial Variational autoencoder with Guided Attention (CAVGA), which localizes the anomaly with a convolutional latent variable to preserve the spatial information. In the unsupervised setting, we propose an attention expansion loss, where we encourage CAVGA to focus on all normal regions in the image without using any anomalous training image. Furthermore, using only 2% anomalous images in the weakly supervised setting we propose a complementary guided attention loss, where we encourage the normal attention to focus on all normal regions while minimizing the regions covered by the anomalous attention in the normal image. CAVGA outperforms the state-of-the-art (SOTA) anomaly detection methods on the MNIST, CIFAR-10, Fashion-MNIST, MVTec Anomaly Detection (MVTAD), and modified ShanghaiTech Campus (mSTC) datasets. CAVGA also outperforms the SOTA anomaly localization methods on the MVTAD and mSTC datasets.

show abstract

Anomaly Detection in Nanofibrous Materials by CNN-Based Self-Similarity

Cited by 246 publications

References 52 publications

Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders

Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders

Constrained Image Splicing Detection and Localization With Attention-Aware Encoder-Decoder and Atrous Convolution

Attention Guided Anomaly Localization in Images

Contact Info

Product

Resources

About