XiaoiceSing is a singing voice synthesis (SVS) system that aims at generating 48kHz singing voices. However, the mel-spectrogram generated by it is over-smoothing in middle- and high-frequency areas due to no special design for modeling the details of these parts. In this paper, we propose XiaoiceSing2, which can generate the details of middle- and high-frequency parts to better construct the full-band mel-spectrogram. Specifically, in order to alleviate this problem, XiaoiceSing2 adopts a generative adversarial network (GAN), which consists of a FastSpeech-based generator and a multi-band discriminator. We improve the feed-forward Transformer (FFT) block by adding multiple residual convolutional blocks in parallel with the self-attention block to balance the local and global features. The multi-band discriminator contains three sub-discriminators responsible for low-, middle-, and high-frequency parts of the mel-spectrogram, respectively. Each sub-discriminator is composed of several segment discriminators (SD) and detail discriminators (DD) to distinguish the audio from different aspects. The experiment on our internal 48kHz singing voice dataset shows XiaoiceSing2 significantly improves the quality of the singing voice over XiaoiceSing.
Transcranial focused ultrasound (tFUS) is a therapeutic ultrasound method that focuses sound through the skull to a small region noninvasively and often under MRI guidance. CT imaging is used to estimate the acoustic properties that vary between individual skulls to enable effective focusing during tFUS procedures, exposing patients to potentially harmful radiation. A method to estimate acoustic parameters in the skull without the need for CT would be desirable. Here, we synthesized CT images from routinely acquired T1-weighted MRI by using a 3D patch-based conditional generative adversarial network (cGAN) and evaluated the performance of synthesized CT (sCT) images for treatment planning with tFUS. We compared the performance of sCT to real CT (rCT) images for tFUS planning using Kranion and simulations using the acoustic toolbox, k-Wave. Simulations were performed for 3 tFUS scenarios: 1) no aberration correction, 2) correction with phases calculated from Kranion, and 3) phase shifts calculated from time-reversal. From Kranion, skull density ratio, skull thickness, and number of active elements between rCT and sCT had Pearson's Correlation Coefficients of 0.94, 0.92, and 0.98, respectively. Among 20 targets, differences in simulated peak pressure between rCT and sCT were largest without phase correction (12.4$\pm$8.1%) and smallest with Kranion phases (7.3$\pm$6.0%). The distance between peak focal locations between rCT and sCT was less than 1.3 mm for all simulation cases. Real and synthetically generated skulls had comparable image similarity, skull measurements, and acoustic simulation metrics. Our work demonstrates the feasibility of replacing real CTs with the MR-synthesized CT for tFUS planning. Source code and a docker image with the trained model are available at https://github.com/han-liu/SynCT_TcMRgFUS
Generative adversarial network (GAN) continues to be a popular research direction due to its high generation quality. It is observed that many state-of-the-art GANs generate samples that are more similar to the training set than a holdout testing set from the same distribution, hinting some training samples are implicitly memorized in these models. This memorization behavior is unfavorable in many applications that demand the generated samples to be sufficiently distinct from known samples. Nevertheless, it is unclear whether it is possible to reduce memorization without compromising the generation quality. In this paper, we propose memorization rejection, a training scheme that rejects generated samples that are near-duplicates of training samples during training. Our scheme is simple, generic and can be directly applied to any GAN architecture. Experiments on multiple datasets and GAN models validate that memorization rejection effectively reduces training sample memorization, and in many cases does not sacrifice the generation quality. Code to reproduce the experiment results can be found at $\texttt{https://github.com/jybai/MRGAN}$.
General movement assessment (GMA) of infant movement videos (IMVs) is an effective method for early detection of cerebral palsy (CP) in infants. We demonstrate in this paper that end-to-end trainable neural networks for image sequence recognition can be applied to achieve good results in GMA, and more importantly, augmenting raw video with infant body parsing and pose estimation information can significantly improve performance. To solve the problem of efficiently utilizing partially labeled IMVs for body parsing, we propose a semi-supervised model, termed SiamParseNet (SPN), which consists of two branches, one for intra-frame body parts segmentation and another for inter-frame label propagation. During training, the two branches are jointly trained by alternating between using input pairs of only labeled frames and input of both labeled and unlabeled frames. We also investigate training data augmentation by proposing a factorized video generative adversarial network (FVGAN) to synthesize novel labeled frames for training. When testing, we employ a multi-source inference mechanism, where the final result for a test frame is either obtained via the segmentation branch or via propagation from a nearby key frame. We conduct extensive experiments for body parsing using SPN on two infant movement video datasets, where SPN coupled with FVGAN achieves state-of-the-art performance. We further demonstrate that SPN can be easily adapted to the infant pose estimation task with superior performance. Last but not least, we explore the clinical application of our method for GMA. We collected a new clinical IMV dataset with GMA annotations, and our experiments show that SPN models for body parsing and pose estimation trained on the first two datasets generalize well to the new clinical dataset and their results can significantly boost the CRNN-based GMA prediction performance.
The issue of image haze removal has attracted wide attention in recent years. However, most existing haze removal methods cannot restore the scene with clear blue sky, since the color and texture information of the object in the original haze image is insufficient. To remedy this, we propose a cycle generative adversarial network to construct a novel end-to-end image dehaze model. We adopt outdoor image datasets to train our model, which includes a set of real-world unpaired image dataset and a set of paired image dataset to ensure that the generated images are close to the real scene. Based on the cycle structure, our model adds four different kinds of loss function to constrain the effect including adversarial loss, cycle consistency loss, photorealism loss and paired L1 loss. These four constraints can improve the overall quality of such degraded images for better visual appeal and ensure reconstruction of images to keep from distortion. The proposed model could remove the haze of images and also restore the sky of images to be clean and blue (like captured in a sunny weather).
In order to protect the privacy of speech data, speaker anonymization aims for hiding the identity of a speaker by changing the voice in speech recordings. This typically comes with a privacy-utility trade-off between protection of individuals and usability of the data for downstream applications. One of the challenges in this context is to create non-existent voices that sound as natural as possible. In this work, we propose to tackle this issue by generating speaker embeddings using a generative adversarial network with Wasserstein distance as cost function. By incorporating these artificial embeddings into a speech-to-text-to-speech pipeline, we outperform previous approaches in terms of privacy and utility. According to standard objective metrics and human evaluation, our approach generates intelligible and content-preserving yet privacy-protecting versions of the original recordings.
We propose AudioStyleGAN (ASGAN), a new generative adversarial network (GAN) for unconditional speech synthesis. As in the StyleGAN family of image synthesis models, ASGAN maps sampled noise to a disentangled latent vector which is then mapped to a sequence of audio features so that signal aliasing is suppressed at every layer. To successfully train ASGAN, we introduce a number of new techniques, including a modification to adaptive discriminator augmentation to probabilistically skip discriminator updates. ASGAN achieves state-of-the-art results in unconditional speech synthesis on the Google Speech Commands dataset. It is also substantially faster than the top-performing diffusion models. Through a design that encourages disentanglement, ASGAN is able to perform voice conversion and speech editing without being explicitly trained to do so. ASGAN demonstrates that GANs are still highly competitive with diffusion models. Code, models, samples: https://github.com/RF5/simple-asgan/.
The goal of semantic image synthesis is to generate photo-realistic images from semantic label maps. It is highly relevant for tasks like content generation and image editing. Current state-of-the-art approaches, however, still struggle to generate realistic objects in images at various scales. In particular, small objects tend to fade away and large objects are often generated as collages of patches. In order to address this issue, we propose a Dual Pyramid Generative Adversarial Network (DP-GAN) that learns the conditioning of spatially-adaptive normalization blocks at all scales jointly, such that scale information is bi-directionally used, and it unifies supervision at different scales. Our qualitative and quantitative results show that the proposed approach generates images where small and large objects look more realistic compared to images generated by state-of-the-art methods.
Botnet detectors based on machine learning are potential targets for adversarial evasion attacks. Several research works employ adversarial training with samples generated from generative adversarial nets (GANs) to make the botnet detectors adept at recognising adversarial evasions. However, the synthetic evasions may not follow the original semantics of the input samples. This paper proposes a novel GAN model leveraged with deep reinforcement learning (DRL) to explore semantic aware samples and simultaneously harden its detection. A DRL agent is used to attack the discriminator of the GAN that acts as a botnet detector. The discriminator is trained on the crafted perturbations by the agent during the GAN training, which helps the GAN generator converge earlier than the case without DRL. We name this model RELEVAGAN, i.e. ["relive a GAN" or deep REinforcement Learning-based Evasion Generative Adversarial Network] because, with the help of DRL, it minimises the GAN's job by letting its generator explore the evasion samples within the semantic limits. During the GAN training, the attacks are conducted to adjust the discriminator weights for learning crafted perturbations by the agent. RELEVAGAN does not require adversarial training for the ML classifiers since it can act as an adversarial semantic-aware botnet detection model. Code will be available at https://github.com/rhr407/RELEVAGAN.
Conditional generation of time-dependent data is a task that has much interest, whether for data augmentation, scenario simulation, completing missing data, or other purposes. Recent works proposed a Transformer-based Time series generative adversarial network (TTS-GAN) to address the limitations of recurrent neural networks. However, this model assumes a unimodal distribution and tries to generate samples around the expectation of the real data distribution. One of its limitations is that it may generate a random multivariate time series; it may fail to generate samples in the presence of multiple sub-components within an overall distribution. One could train models to fit each sub-component separately to overcome this limitation. Our work extends the TTS-GAN by conditioning its generated output on a particular encoded context allowing the use of one model to fit a mixture distribution with multiple sub-components. Technically, it is a conditional generative adversarial network that models realistic multivariate time series under different types of conditions, such as categorical variables or multivariate time series. We evaluate our model on UniMiB Dataset, which contains acceleration data following the XYZ axes of human activities collected using Smartphones. We use qualitative evaluations and quantitative metrics such as Principal Component Analysis (PCA), and we introduce a modified version of the Frechet inception distance (FID) to measure the performance of our model and the statistical similarities between the generated and the real data distributions. We show that this transformer-based CGAN can generate realistic high-dimensional and long data sequences under different kinds of conditions.
Previous studies have shown that commonly studied (vanilla) implementations of touch-based continuous authentication systems (V-TCAS) are susceptible to active adversarial attempts. This study presents a novel Generative Adversarial Network assisted TCAS (G-TCAS) framework and compares it to the V-TCAS under three active adversarial environments viz. Zero-effort, Population, and Random-vector. The Zero-effort environment was implemented in two variations viz. Zero-effort (same-dataset) and Zero-effort (cross-dataset). The first involved a Zero-effort attack from the same dataset, while the second used three different datasets. G-TCAS showed more resilience than V-TCAS under the Population and Random-vector, the more damaging adversarial scenarios than the Zero-effort. On average, the increase in the false accept rates (FARs) for V-TCAS was much higher (27.5% and 21.5%) than for G-TCAS (14% and 12.5%) for Population and Random-vector attacks, respectively. Moreover, we performed a fairness analysis of TCAS for different genders and found TCAS to be fair across genders. The findings suggest that we should evaluate TCAS under active adversarial environments and affirm the usefulness of GANs in the TCAS pipeline.
Previous virtual try-on methods usually focus on aligning a clothing item with a person, limiting their ability to exploit the complex pose, shape and skin color of the person, as well as the overall structure of the clothing, which is vital to photo-realistic virtual try-on. To address this potential weakness, we propose a fill in fabrics (FIFA) model, a self-supervised conditional generative adversarial network based framework comprised of a Fabricator and a unified virtual try-on pipeline with a Segmenter, Warper and Fuser. The Fabricator aims to reconstruct the clothing image when provided with a masked clothing as input, and learns the overall structure of the clothing by filling in fabrics. A virtual try-on pipeline is then trained by transferring the learned representations from the Fabricator to Warper in an effort to warp and refine the target clothing. We also propose to use a multi-scale structural constraint to enforce global context at multiple scales while warping the target clothing to better fit the pose and shape of the person. Extensive experiments demonstrate that our FIFA model achieves state-of-the-art results on the standard VITON dataset for virtual try-on of clothing items, and is shown to be effective at handling complex poses and retaining the texture and embroidery of the clothing.
Synthetic data generated by generative models can enhance the performance and capabilities of data-hungry deep learning models in medical imaging. However, there is (1) limited availability of (synthetic) datasets and (2) generative models are complex to train, which hinders their adoption in research and clinical applications. To reduce this entry barrier, we propose medigan, a one-stop shop for pretrained generative models implemented as an open-source framework-agnostic Python library. medigan allows researchers and developers to create, increase, and domain-adapt their training data in just a few lines of code. Guided by design decisions based on gathered end-user requirements, we implement medigan based on modular components for generative model (i) execution, (ii) visualisation, (iii) search & ranking, and (iv) contribution. The library's scalability and design is demonstrated by its growing number of integrated and readily-usable pretrained generative models consisting of 21 models utilising 9 different Generative Adversarial Network architectures trained on 11 datasets from 4 domains, namely, mammography, endoscopy, x-ray, and MRI. Furthermore, 3 applications of medigan are analysed in this work, which include (a) enabling community-wide sharing of restricted data, (b) investigating generative model evaluation metrics, and (c) improving clinical downstream tasks. In (b), extending on common medical image synthesis assessment and reporting standards, we show Fr\'echet Inception Distance variability based on image normalisation and radiology-specific feature extraction.
Under-display camera (UDC) provides an elegant solution for full-screen smartphones. However, UDC captured images suffer from severe degradation since sensors lie under the display. Although this issue can be tackled by image restoration networks, these networks require large-scale image pairs for training. To this end, we propose a modular network dubbed MPGNet trained using the generative adversarial network (GAN) framework for simulating UDC imaging. Specifically, we note that the UDC imaging degradation process contains brightness attenuation, blurring, and noise corruption. Thus we model each degradation with a characteristic-related modular network, and all modular networks are cascaded to form the generator. Together with a pixel-wise discriminator and supervised loss, we can train the generator to simulate the UDC imaging degradation process. Furthermore, we present a Transformer-style network named DWFormer for UDC image restoration. For practical purposes, we use depth-wise convolution instead of the multi-head self-attention to aggregate local spatial information. Moreover, we propose a novel channel attention module to aggregate global information, which is critical for brightness recovery. We conduct evaluations on the UDC benchmark, and our method surpasses the previous state-of-the-art models by 1.23 dB on the P-OLED track and 0.71 dB on the T-OLED track, respectively.
In this work, we further develop the conformer-based metric generative adversarial network (CMGAN) model for speech enhancement (SE) in the time-frequency (TF) domain. This paper builds on our previous work but takes a more in-depth look by conducting extensive ablation studies on model inputs and architectural design choices. We rigorously tested the generalization ability of the model to unseen noise types and distortions. We have fortified our claims through DNS-MOS measurements and listening tests. Rather than focusing exclusively on the speech denoising task, we extend this work to address the dereverberation and super-resolution tasks. This necessitated exploring various architectural changes, specifically metric discriminator scores and masking techniques. It is essential to highlight that this is among the earliest works that attempted complex TF-domain super-resolution. Our findings show that CMGAN outperforms existing state-of-the-art methods in the three major speech enhancement tasks: denoising, dereverberation, and super-resolution. For example, in the denoising task using the Voice Bank+DEMAND dataset, CMGAN notably exceeded the performance of prior models, attaining a PESQ score of 3.41 and an SSNR of 11.10 dB. Audio samples and CMGAN implementations are available online.
Although recent gaze estimation methods lay great emphasis on attentively extracting gaze-relevant features from facial or eye images, how to define features that include gaze-relevant components has been ambiguous. This obscurity makes the model learn not only gaze-relevant features but also irrelevant ones. In particular, it is fatal for the cross-dataset performance. To overcome this challenging issue, we propose a gaze-aware analytic manipulation method, based on a data-driven approach with generative adversarial network inversion's disentanglement characteristics, to selectively utilize gaze-relevant features in a latent code. Furthermore, by utilizing GAN-based encoder-generator process, we shift the input image from the target domain to the source domain image, which a gaze estimator is sufficiently aware. In addition, we propose gaze distortion loss in the encoder that prevents the distortion of gaze information. The experimental results demonstrate that our method achieves state-of-the-art gaze estimation accuracy in a cross-domain gaze estimation tasks. This code is available at https://github.com/leeisack/LatentGaze/.
The goal of this work is to help mitigate the already existing gender wage gap by supplying unbiased job recommendations based on resumes from job seekers. We employ a generative adversarial network to remove gender bias from word2vec representations of 12M job vacancy texts and 900k resumes. Our results show that representations created from recruitment texts contain algorithmic bias and that this bias results in real-world consequences for recommendation systems. Without controlling for bias, women are recommended jobs with significantly lower salary in our data. With adversarially fair representations, this wage gap disappears, meaning that our debiased job recommendations reduce wage discrimination. We conclude that adversarial debiasing of word representations can increase real-world fairness of systems and thus may be part of the solution for creating fairness-aware recommendation systems.
Deep reinforcement learning has demonstrated superhuman performance in complex decision-making tasks, but it struggles with generalization and knowledge reuse - key aspects of true intelligence. This article introduces a novel approach that modifies Cycle Generative Adversarial Networks specifically for reinforcement learning, enabling effective one-to-one knowledge transfer between two tasks. Our method enhances the loss function with two new components: model loss, which captures dynamic relationships between source and target tasks, and Q-loss, which identifies states significantly influencing the target decision policy. Tested on the 2-D Atari game Pong, our method achieved 100% knowledge transfer in identical tasks and either 100% knowledge transfer or a 30% reduction in training time for a rotated task, depending on the network architecture. In contrast, using standard Generative Adversarial Networks or Cycle Generative Adversarial Networks led to worse performance than training from scratch in the majority of cases. The results demonstrate that the proposed method ensured enhanced knowledge generalization in deep reinforcement learning.
Clinical screening with low-quality fundus images is challenging and significantly leads to misdiagnosis. This paper addresses the issue of improving the retinal image quality and vessel segmentation through retinal image restoration. More specifically, a cycle-consistent generative adversarial network (CycleGAN) with a convolution block attention module (CBAM) is used for retinal image restoration. A modified UNet is used for retinal vessel segmentation for the restored retinal images (CBAM-UNet). The proposed model consists of two generators and two discriminators. Generators translate images from one domain to another, i.e., from low to high quality and vice versa. Discriminators classify generated and original images. The retinal vessel segmentation model uses downsampling, bottlenecking, and upsampling layers to generate segmented images. The CBAM has been used to enhance the feature extraction of these models. The proposed method does not require paired image datasets, which are challenging to produce. Instead, it uses unpaired data that consists of low- and high-quality fundus images retrieved from publicly available datasets. The restoration performance of the proposed method was evaluated using full-reference evaluation metrics, e.g., peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). The retinal vessel segmentation performance was compared with the ground-truth fundus images. The proposed method can significantly reduce the degradation effects caused by out-of-focus blurring, color distortion, low, high, and uneven illumination. Experimental results show the effectiveness of the proposed method for retinal image restoration and vessel segmentation.
Single-Image Super-Resolution can support robotic tasks in environments where a reliable visual stream is required to monitor the mission, handle teleoperation or study relevant visual details. In this work, we propose an efficient Generative Adversarial Network model for real-time Super-Resolution, called EdgeSRGAN (code available at https://github.com/PIC4SeR/EdgeSRGAN). We adopt a tailored architecture of the original SRGAN and model quantization to boost the execution on CPU and Edge TPU devices, achieving up to 200 fps inference. We further optimize our model by distilling its knowledge to a smaller version of the network and obtain remarkable improvements compared to the standard training approach. Our experiments show that our fast and lightweight model preserves considerably satisfying image quality compared to heavier state-of-the-art models. Finally, we conduct experiments on image transmission with bandwidth degradation to highlight the advantages of the proposed system for mobile robotic applications.