IntereStyle: Encoding an Interest Region for Robust StyleGAN Inversion

Generative models rely on the idea that data can be represented in terms of latent variables which are uncorrelated by definition. Lack of correlation among the latent variable support is important because it suggests that the latent-space manifold is simpler to understand and manipulate than the real-space representation. Many types of generative model are used in deep learning, e.g., variational autoencoders (VAEs) and generative adversarial networks (GANs). Based on the idea that the latent space behaves like a vector space Radford et al. (2015), we ask whether we can expand the latent space representation of our data elements in terms of an orthonormal basis set. Here we propose a method to build a set of linearly independent vectors in the latent space of a trained GAN, which we call quasi-eigenvectors. These quasi-eigenvectors have two key properties: i) They span the latent space, ii) A set of these quasi-eigenvectors map to each of the labeled features one-to-one. We show that in the case of the MNIST image data set, while the number of dimensions in latent space is large by design, 98% of the data in real space map to a sub-domain of latent space of dimensionality equal to the number of labels. We then show how the quasi-eigenvectors can be used for Latent Spectral Decomposition (LSD). We apply LSD to denoise MNIST images. Finally, using the quasi-eigenvectors, we construct rotation matrices in latent space which map to feature transformations in real space. Overall, from quasi-eigenvectors we gain insight regarding the latent space topology.

In the field of deep learning, the generative model did not attract much attention until GANs (generative adversarial networks) appeared. In 2014, Google’s Ian Goodfellow proposed a generative model called GANs. GANs use different structures and objective functions from the existing generative model. For example, GANs use two neural networks: a generator that creates a realistic image, and a discriminator that distinguishes whether the input is real or synthetic. If there are no problems in the training process, GANs can generate images that are difficult even for experts to distinguish in terms of authenticity. Currently, GANs are the most researched subject in the field of computer vision, which deals with the technology of image style translation, synthesis, and generation, and various models have been unveiled. The issues raised are also improving one by one. In image synthesis, BEGAN (Boundary Equilibrium Generative Adversarial Network), which outperforms the previously announced GANs, learns the latent space of the image, while balancing the generator and discriminator. Nonetheless, BEGAN also has a mode collapse wherein the generator generates only a few images or a single one. Although BEGAN-CS (Boundary Equilibrium Generative Adversarial Network with Constrained Space), which was improved in terms of loss function, was introduced, it did not solve the mode collapse. The discriminator structure of BEGAN-CS is AE (AutoEncoder), which cannot create a particularly useful or structured latent space. Compression performance is not good either. In this paper, this characteristic of AE is considered to be related to the occurrence of mode collapse. Thus, we used VAE (Variational AutoEncoder), which added statistical techniques to AE. As a result of the experiment, the proposed model did not cause mode collapse but converged to a better state than BEGAN-CS.

Extra-Tropical Cyclones are major systems ruling and influencing the atmospheric structure at mid-latitudes. They are characterised by strong winds and heavy precipitation, and can cause considerable storm surges potentially devastating for coastal regions. The availability of historical observations of the extreme events caused by intense ETCs are rather limited, hampering risk evaluation. Increasing the amount of significant data available would substantially help several fields of analysis influenced by these events, such as coastal management, agricultural production, energy distribution, air and maritime transportation, and risk assessment and management.Here, we address the possibility of generating synthetic ETC atmospheric fields of mean sea level pressure, wind speed, and precipitation in the North Atlantic by training a Generative Adversarial Network (GAN). The purpose of GANs is to learn the distribution of a training set based on a game theoretic scenario where two networks compete against each other, the generator and the discriminator. The former is trained to generate synthetic examples that are plausible and resemble the real ones. The input of the generator is a vector of random Gaussian values, whose domain is known as the “latent space”. The discriminator learns to distinguish whether an example comes from the dataset distribution. The competition set by the game-theoretic approach improves the network until the counterfeits are indistinguishable from the originals.To train the GAN, we use atmospheric fields extracted from the ERA5 reanalysis dataset in the geographic domain with boundaries 0°- 90°N, 70°W - 20°E and for the period 1st January 1979 - 1st January 2020. We analyse the generated samples’ histograms, the samples’ average fields, the Wasserstein distance and the Kullback-Leibler divergence between the generated samples and the test set distributions. Results show that the generative model has learned the distribution of the values of the atmospheric fields and the general spatial trends of the atmosphere in the domain. To evaluate better the atmospheric structure learned by the network, we perform linear and spherical interpolations in the latent space. Specifically, we consider four cyclones and compare the frames of their tracks to those of the synthetic tracks generated by interpolation. The interpolated tracks show interesting features consistent with the original tracks. These findings suggest that GANs can learn meaningful representations of the ETCs’ fields, encouraging further investigations to model the tracks’ temporal evolution.

In this paper, a flexible region of interest (ROI) image coding method so-called partial block bitplane shift (PBBShift) is presented for JPEG2000. The new method offers two strategies - bitplane block scaling (BBS) and partial bitplane shift (PBS), which support arbitrarily shaped single or multiple ROI. In BBS, all bitplanes of ROI and background (BG) are divide into four parts - most significant ROI (MSR), general significant ROI (GSR), general significant BG (GSB), and least significant BG (LSB). In PBS, all bitplanes of MSR are shifted up over all residual bitplanes. The bitplanes in LSB are shifted down below all bitplanes. The bitplanes of GSR and GSB are not shifted. Experimental result shows that the presented method can flexibly adjust the compression quality between ROI and background by selecting an arbitrary scaling value without any ROI shape information. Additionally, PBBShift can efficiently code multiple ROIs with different degrees of interest in an image.

Breast cancer is one of the most common cancer types especially met in women. The number of breast cancer patients increases every year. Thus, to detect breast cancer at its early stages gains importance. Breast region detection is the pioneering step of breast cancer diagnosis researches performed via image processing techniques. The performance of computer-aided breast cancer diagnosis systems can be improved by exactly determining the breast region of interest. In this study, the goal is to determine a region of interest for breast MR images, in which one or more lesion can appear. The achieved region includes two breasts and lymph nodes. The proposed region of interest detection system is fully automatic and it utilizes several image processing techniques. At first, the local adaptive thresholding technique is applied to the noise-filtered grey level breast magnetic resonance images taken with ethical permissions from Sakarya Education and Research Hospital. After adaptive thresholding, connected component analysis is performed to exclude extra structures around the breast region as thorax area. This analysis selects the largest area in the binary image which corresponds to a gyrate region including breast area and lymph nodes over the backbone. Then, the integral of horizontal projection is calculated to determine an optimum horizontal line that allows setting the region of interest apart. In the following step, sternum midpoint is detected to separate the right breast from the left one. Finally, a masking operation is applied to get corresponding right and left breast regions in the original MR image. To evaluate the performance of the proposed study, the results of automatic region of interest detection system are compared with the manual region of interest selection performed by an expert radiologist. Dice similarity coefficient and Jaccard coefficient are used as performance criteria. According to the results, the proposed system can detect region of interest for computer-aided breast cancer diagnosis researches, exactly.

Deep generative models are promising in detecting novel cyber-physical attacks, mitigating the vulnerability of Cyber-physical systems (CPSs) without relying on labeled information. Nonetheless, these generative models face challenges in identifying attack behaviors that closely resemble normal data, or deviate from the normal data distribution but are in close proximity to the manifold of the normal cluster in latent space. To tackle this problem, this article proposes a novel unsupervised dual variational generative adversarial model named MST-DVGAN, to perform anomaly detection in multivariate time series data for CPS security. The central concept is to enhance the model's discriminative capability by widening the distinction between reconstructed abnormal samples and their normal counterparts. Specifically, we propose an augmented module by imposing contrastive constraints on the reconstruction process to obtain a more compact embedding. Then, by exploiting the distribution property and modeling the normal patterns of multivariate time series, a variational autoencoder is introduced to force the generative adversarial network (GAN) to generate diverse samples. Furthermore, two augmented loss functions are designed to extract essential characteristics in a self-supervised manner through mutual guidance between the augmented samples and original samples. Finally, a specific feature center loss is introduced for the generator network to enhance its stability. Empirical experiments are conducted on three public datasets, namely SWAT, WADI and NSL_KDD. Comparing with the state-of-the-art methods, the evaluation results show that the proposed MTS-DVGAN is more stable and can achieve consistent performance improvement.

Auto-Encoder (AE) based neural generative frameworks model the joint-distribution between the data and the latent space using an Encoder-Decoder pair, with regularization imposed in terms of a prior over the latent space. Despite their advantages, such as stability in training, efficient inference, the performance of AE based models has not reached the superior standards of the other generative models such as Generative Adversarial Networks (GANs). Motivated by this, we examine the effect of the latent prior on the generation quality of deterministic AE models in this paper. Specifically, we consider the class of Generative AE models with deterministic Encoder-Decoder pair (such as Wasserstein Auto-Encoder (WAE), Adversarial Auto-Encoder (AAE)), and show that having a fixed prior distribution, \textit{a priori}, oblivious to the dimensionality of the `true' latent space, will lead to the infeasibility of the optimization problem considered. As a remedy to the issue mentioned above, we introduce an additional state space in the form of flexibly learnable latent priors, in the optimization objective of WAE/AAE. Additionally, we employ a latent-space interpolation based smoothing scheme to address the non-smoothness that may arise from highly flexible priors. We show the efficacy of our proposed models, called FlexAE and FlexAE-SR, through several experiments on multiple datasets, and demonstrate that FlexAE-SR is the new state-of-the-art for the AE based generative models in terms of generation quality as measured by several metrics such as Fr\'echet Inception Distance, Precision/Recall score. Code for our paper is available at: \url{https://github.com/dair-iitd/FlexAE}

Generative models such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) play an increasingly important role in medical image analysis. The latent spaces of these models often show semantically meaningful directions corresponding to human-interpretable image transformations. However, until now, their exploration for medical images has been limited due to the requirement of supervised data. Several methods for unsupervised discovery of interpretable directions in GAN latent spaces have shown interesting results on natural images. This work explores the potential of applying these techniques on medical images by training a GAN and a VAE on thoracic CT scans and using an unsupervised method to discover interpretable directions in the resulting latent space. We find several directions corresponding to non-trivial image transformations, such as rotation or breast size. Furthermore, the directions show that the generative models capture 3D structure despite being presented only with 2D data. The results show that unsupervised methods to discover interpretable directions in GANs generalize to VAEs and can be applied to medical images. This opens a wide array of future work using these methods in medical image analysis. The code and animations of the discovered directions are available online at https://github.com/julschoen/Latent-Space-Exploration-CT.KeywordsGenerative modelsUnsupervised learningInterpretabilityCT

In breast X-ray imaging, breast texture has been characterized by a radial noise power spectrum (NPS) that has an inverse power-law shape with exponent β. The technique to estimate the radial power-law coefficient β is typically based on averaging 2-dimensional noise power spectra (NPS), calculated from partly overlapping image regions each weighted by a suitable window function. The linear regression applied over a selected frequency range to the logarithm of the 1- dimensional NPS as a function of the logarithm of the radial frequencies, gives β. For each step in this process, several alternative techniques have been proposed. This paper investigates the effect of image region of interest (ROI) size, image data windowing and alternative ways to determine radial frequency in terms of bias, variance and root mean square error (RMSE) in the estimated β. The effects of these three factors were analytically derived and evaluated using synthetic images with known β varying from 1 to 4 to cover the range of textures encountered in 2D and 3D breast X-ray imaging. Our results indicate that the RMSE in estimated β is smallest when the ROIs are multiplied with an appropriate window function and either no radial averaging or radial averaging with small frequency bins is applied. The ROI size yielding the smallest RMSE depends on several factors and needs to be validated with numerical simulations. In clinical practice however, there might be a need to compromise in the choice of the ROI size to balance between the RMSE magnitudes inherent to the applied β estimation technique and encompass the breast texture range so as to obtain an accurate shape of the NPS. When using 2.56 cm x 2.56 cm ROI sizes, applying a 2D Hann window and no radial frequency averaging, the RMSE in the estimated β ranges from 0.04 to 0.1 for true β values equal to 1 and 4. While many subtleties in real images were not modeled to simplify the mathematics in deriving our results, this work is illustrative in demonstrating the limits of commonly used algorithm steps to estimate accurate β values.

Image super-resolution is a research endeavour that has gained notoriety in computer vision. The research goal is to increase the spatial dimensions of an image using corresponding low-resolution and high-resolution image pairs to enhance the perceptual quality. The challenge of maintaining such perceptual quality lies in developing appropriate algorithms that learn to reconstruct higher-quality images from their lower-resolution counterparts. Recent methods employ deep learning algorithms to reconstruct textural details prevalent in low-resolution images. Since corresponding image pairs are non-trivial to collect, researchers attempt super-resolution by creating synthetic low-resolution representations of high-resolution images. Unfortunately, such methods employ ineffective downscaling operations to achieve synthetic low-resolution images. These methods fail to generalize well on real-world images that may suffer different degradations. A different angle is offered to solve the task of image super-resolution by investigating the plausibility of learning the degradational operation using generative adversarial networks. A two-stage generative adversarial network along with two architectural variations is proposed to solve the task of real-world super-resolution from low-resolution images with unknown degradations. It is demonstrated that learning to downsample images in a weakly supervised manner is an impactful and viable approach for super-resolution.

Facial expressions in infrared imaging have been introduced to solve the problem of illumination, which is an integral constituent of visual imagery. The paper investigates facial skin temperature distribution on mixed thermal facial expressions of our created face database where six are basic expressions and rest 12 are a mixture of those basic expressions. Temperature analysis has been performed on three facial regions of interest (ROIs); periorbital, supraorbital and mouth. Temperature variability of the ROIs in different expressions has been measured using statistical parameters. The temperature variation measurement in ROIs of a particular expression corresponds to a vector, which is later used in recognition of mixed facial expressions. Investigations show that facial features in mixed facial expressions can be characterized by positive emotion induced facial features and negative emotion induced facial features. Supraorbital is a useful facial region that can differentiate basic expressions from mixed expressions. Analysis and interpretation of mixed expressions have been conducted with the help of box and whisker plot. Facial region containing mixture of two expressions is generally less temperature inducing than corresponding facial region containing basic expressions.

Image dehazing has always been a challenging topic in image processing. The development of deep learning methods, especially the generative adversarial networks (GAN), provides a new way for image dehazing. In recent years, many deep learning methods based on GAN have been applied to image dehazing. However, GAN has two problems in image dehazing. Firstly, For haze image, haze not only reduces the quality of the image but also blurs the details of the image. For GAN network, it is difficult for the generator to restore the details of the whole image while removing the haze. Secondly, GAN model is defined as a minimax problem, which weakens the loss function. It is difficult to distinguish whether GAN is making progress in the training process. Therefore, we propose a guided generative adversarial dehazing network (GGADN). Different from other generation adversarial networks, GGADN adds a guided module on the generator. The guided module verifies the network of each layer of the generator. At the same time, the details of the map generated by each layer are strengthened. Network training is based on the pre-trained VGG feature model and L1-regularized gradient prior which is developed by new loss function parameters. From the dehazing results of synthetic images and real images, the proposed method is better than the state-of-the-art dehazing methods.

Precise lesion analysis to detect diabetic retinopathy using Generative Adversarial Network(GAN) and Mask-RCNN

Learning based single image super-resolution (SISR) for real-world images has been an active research topic yet a challenging task, due to the lack of paired low-resolution (LR) and high-resolution (HR) training images. Most of the existing unsupervised real-world SISR methods adopt a two-stage training strategy by synthesizing realistic LR images from their HR counterparts first, then training the super-resolution (SR) models in a supervised manner. However, the training of image degradation and SR models in this strategy are separate, ignoring the inherent mutual dependency between downscaling and its inverse upscaling process. Additionally, the ill-posed nature of image degradation is not fully considered. In this paper, we propose an image downscaling and SR model dubbed as SDFlow, which simultaneously learns a bidirectional many-to-many mapping between real-world LR and HR images unsupervisedly. The main idea of SDFlow is to decouple image content and degradation information in the latent space, where content information distribution of LR and HR images is matched in a common latent space. Degradation information of the LR images and the high-frequency information of the HR images are fitted to an easy-to-sample conditional distribution. Experimental results on real-world image SR datasets indicate that SDFlow can generate diverse realistic LR and SR images both quantitatively and qualitatively.

As an important display mode of underwater environments, the sonar image has limitations on the resolution, which often leads to problems with low resolution of underwater objects. Therefore, the image super-resolution algorithm is needed to transform the images from low-resolution to high-resolution. It can improve the visual effect and contribute to subsequent processing such as 3D reconstruction and object recognition. This paper proposes a method for sonar image super-resolution based on generative adversarial networks (GAN). By comparing the super-resolution effects of various interpolation and convolutional neural network algorithms on sonar images, a Residual-in-Residual Dense Block network is employed as the generator of GAN since it has the low distortion and high perceptual quality. Because the sonar image training set does not have enough data, the generator utilizes the transfer learning on the sonar images to produce an optimized network model which is more suitable for super-resolution of sonar image. The VGG19 network is employed as the discriminator. In addition, the perceptual loss is introduced into the loss function of $$\hbox {S}^2\hbox {RGAN}$$ to further improve the perceptual quality of super-resolution images. The experimental results indicate that the proposed $$\hbox {S}^2\hbox {RGAN}$$ shows excellent performance. The generated super-resolution images of $$\hbox {S}^2\hbox {RGAN}$$ have the remarkable advantages of both lower distortion and higher perceptual quality comparing with other methods. Because $$\hbox {S}^2\hbox {RGAN}$$ focuses more on the reality and overall visual effect of super-resolution sonar images, it is suitable for various underwater situations.