Single-image 3-D reconstruction has long been a challenging problem. Recent deep learning approaches have been introduced to this 3-D area, but the ability to generate point clouds still remains limited due to inefficient and expensive 3-D representations, the dependency between the output and the number of model parameters, or the lack of a suitable computing operation. In this article, we present a novel deep-learning-based method to reconstruct a point cloud of an object from a single still image. The proposed method can be decomposed into two steps: feature fusion and deformation. The first step extracts both global and point-specific shape features from a 2-D object image, and then injects them into a randomly generated point cloud. In the second step, which is deformation, we introduce a new layer termed as GraphX that considers the interrelationship between points like common graph convolutions but operates on unordered sets. The framework can be applicable to realistic image data with background as we optionally learn a mask branch to segment objects from input images. To complement the quality of point clouds, we further propose an objective function to control the point uniformity. In addition, we introduce different variants of GraphX that cover from best performance to best memory budget. Moreover, the proposed model can generate an arbitrary-sized point cloud, which is the first deep method to do so. Extensive experiments demonstrate that we outperform the existing models and set a new height for different performance metrics in single-image 3-D reconstruction.

Image inpainting methods leverage the similarity of adjacent pixels to create alternative content. However, as the invisible region becomes larger, the pixels completed in the deeper hole are difficult to infer from the surrounding pixel signal, which is more prone to visual artifacts. To help fill this void, we adopt an alternative progressive hole-filling scheme that hierarchically fills the corrupted region in the feature and image spaces. This technique allows us to utilize reliable contextual information of the surrounding pixels, even for large hole samples, and then gradually complete the details as the resolution increases. For a more realistic representation of the completed region, we devise a pixel-wise dense detector. By distinguishing each pixel as whether it is a masked region or not, and passing the gradient to all resolutions, the generator further enhances the potential quality of the compositing. Furthermore, the completed images at different resolutions are then merged using a proposed structure transfer module (STM) that incorporates fine-grained local and coarse-grained global interactions. In this new mechanism, each completed image at the different resolutions attends its closest composition at fine granularity adjacent image and thus can capture the global continuity by interacting both short- and long-range dependencies. By comparing our solutions qualitatively and quantitatively with state-of-the-art methods, we conclude that our model exhibits a significantly improved visual quality, even in the case of large holes.

When coming up with phrases of movement, choreographers all have their habits as they are used to their skilled dance genres. Therefore, they tend to return certain patterns of the dance genres that they are familiar with. What if artificial intelligence could be used to help choreographers blend dance genres by suggesting various dances, and one that matches their choreographic style? Numerous task-specific variants of autoregressive networks have been developed for dance generation. Yet, a serious limitation remains that all existing algorithms can return repeated patterns for a given initial pose sequence, which may be inferior. To mitigate this issue, we propose MNET, a novel and scalable approach that can perform music-conditioned pluralistic dance generation synthesized by multiple dance genres using only a single model. Here, we learn a dancegenre aware latent representation by training a conditional generative adversarial network leveraging Transformer architecture. We conduct extensive experiments on AIST++ along with user studies. Compared to the state-of-the-art methods, our method synthesizes plausible and diverse outputs according to multiple dance genres as well as generates outperforming dance sequences qualitatively and quantitatively.

In a virtual reality (VR) environment, where visual stimuli predominate over other stimuli, the user experiences cybersickness because the balance of the body collapses due to self-motion. Accordingly, the VR experience is accompanied by unavoidable sickness referred to as visually induced motion sickness (VIMS). In this article, our primary purpose is to simultaneously estimate the VIMS score by referring to the content and calculate the temporally induced VIMS sensitivity. To seek our goals, we propose a novel architecture composed of two consecutive networks: 1) neurological representation and 2) spatiotemporal representation. In the first stage, the network imitates and learns the neurological mechanism of motion sickness. In the second stage, the significant feature of the spatial and temporal domains is expressed over the generated frames. After the training procedure, our model can calculate VIMS sensitivity for each frame of the VR content by using the weakly supervised approach for unannotated temporal VIMS scores. Furthermore, we release a massive VR content database. In the experiments, the proposed framework demonstrates excellent performance for VIMS score prediction compared with existing methods, including feature engineering and deep learning-based approaches. Furthermore, we propose a way to visualize the cognitive response to visual stimuli and demonstrate that the induced sickness tends to be activated in a similar tendency, as done in clinical studies.

What if we could interpret the cognitive state of a user while experiencing a virtual reality (VR) and estimate the cognitive state from a visual stimulus? In this paper, we address the above question by developing an electroencephalography (EEG) driven VR cybersickness prediction model. The EEG data has been widely utilized to learn the cognitive representation of brain activity. In the first stage, to fully exploit the advantages of the EEG data, it is transformed into the multi-channel spectrogram which enables to account for the correlation of spectral and temporal coefficient. Then, a convolutional neural network (CNN) is applied to encode the cognitive representation of the EEG spectrogram. In the second stage, we train a cybersickness prediction model on the VR video sequence by designing a Recurrent Neural Network (RNN). Here, the encoded cognitive representation is transferred to the model to train the visual and cognitive features for cybersickness prediction. Through the proposed framework, it is possible to predict the cybersickness level that reflects brain activity automatically. We use 8-channels EEG data to record brain activity while more than 200 subjects experience 44 different VR contents. After rigorous training, we demonstrate that the proposed framework reliably estimates cognitive states without the EEG data. Furthermore, it achieves state-of-the-art performance comparing to existing VR cybersickness prediction models.

In previous quality assessment studies on stereoscopic 3D (S3D) images, researchers have concentrated on deriving manually extracted features which represent the quality of images. These features are based on the human visual system or natural scene statistics, but they have not been revealed as a deterministic function, preventing to guarantee the robustness of features. To solve this problem, we introduce a deep learning method for predicting the quality of S3D images without a reference. A convolutional neural network (CNN) model is trained through two-step learning. First, to overcome the lack of training data, patch-based CNNs are introduced. And then, automatically extracted patch features are pooled into image features. Finally, the trained CNN model parameters are updated iteratively using holistic image labeling, i.e., mean opinion score (MOS). The proposed method represents a significant improvement compared to other no-reference (NR) S3D image quality assessment (IQA) algorithms.

Visual saliency on stereoscopic 3D (S3D) images has been shown to be heavily influenced by image quality. Hence, this dependency is an important factor in image quality prediction, image restoration and discomfort reduction, but it is still very difficult to predict such a nonlinear relation in images. In addition, most algorithms specialized in detecting visual saliency on pristine images may unsurprisingly fail when facing distorted images. In this paper, we investigate a deep learning scheme named Deep Visual Saliency (DeepVS) to achieve a more accurate and reliable saliency predictor even in the presence of distortions. Since visual saliency is influenced by low-level features (contrast, luminance and depth information) from a psychophysical point of view, we propose seven low-level features derived from S3D image pairs and utilize them in the context of deep learning to detect visual attention adaptively to human perception. During analysis, it turns out that the low-level features play a role to extract distortion and saliency information. To construct saliency predictors, we weight and model the human visual saliency through two different network architectures, a regression and a fully convolutional neural networks (CNNs). Our results from thorough experiments confirm that the predicted saliency maps are up to 70 % correlated with human gaze patterns, which emphasize the need for the hand-crafted features as input to deep neural networks in S3D saliency detection.

Most prior approaches to the problem of stereoscopic 3D (S3D) visual discomfort prediction (VDP) have focused on the extraction of perceptually meaningful handcrafted features based on models of visual perception and of natural depth statistics. Towards advancing performance on this problem, we have developed a deep learning based VDP model named Deep Visual Discomfort Predictor (DeepVDP). DeepVDP uses a convolutional neural network (CNN) to learn features that are highly predictive of experienced visual discomfort. Since a large amount of reference data is needed to train a CNN, we develop a systematic way of dividing S3D image into local regions defined as patches, and model a patch-based CNN using two sequential training steps. Since it is very difficult to obtain human opinions on each patch, instead a proxy ground-truth label that is generated by an existing S3D visual discomfort prediction algorithm called 3D-VDP is assigned to each patch. These proxy ground-truth labels are used to conduct the first stage of training the CNN. In the second stage, the automatically learned local abstractions are aggregated into global features via a feature aggregation layer. The learned features are iteratively updated via supervised learning on subjective 3D discomfort scores, which serve as ground-truth labels on each S3D image. The patchbased CNN model that has been pretrained on proxy groundtruth labels is subsequently retrained on true global subjective scores. The global S3D visual discomfort scores predicted by the trained DeepVDP model achieve state-of-the-art performance as compared to previous VDP algorithms.

Previously, no-reference (NR) stereoscopic 3D (S3D) image quality assessment (IQA) algorithms have been limited to the extraction of reliable hand-crafted features based on an understanding of the insufficiently revealed human visual system or natural scene statistics. Furthermore, compared with full-reference (FR) S3D IQA metrics, it is difficult to achieve competitive quality score predictions using the extracted features, which are not optimized with respect to human opinion. To cope with this limitation of the conventional approach, we introduce a novel deep learning scheme for NR S3D IQA in terms of local to global feature aggregation. A deep convolutional neural network (CNN) model is trained in a supervised manner through two-step regression. First, to overcome the lack of training data, local patch-based CNNs are modeled, and the FR S3D IQA metric is used to approximate a reference ground-truth for training the CNNs. The automatically extracted local abstractions are aggregated into global features by inserting an aggregation layer in the deep structure. The locally trained model parameters are then updated iteratively using supervised global labeling, i.e., subjective mean opinion score (MOS). In particular, the proposed deep NR S3D image quality evaluator does not estimate the depth from a pair of S3D images. The S3D image quality scores predicted by the proposed method represent a significant improvement over those of previous NR S3D IQA algorithms. Indeed, the accuracy of the proposed method is competitive with FR S3D IQA metrics, having ~ 91% correlation in terms of MOS.

Conventional stereoscopic 3D (S3D) displays do not provide accommodation depth cues of the 3D image or video contents being viewed. The sense of content depths is thus limited to cues supplied by motion parallax (for 3D video), stereoscopic vergence cues created by presenting left and right views to the respective eyes, and other contextual and perspective depth cues. The absence of accommodation cues can induce two kinds of accommodation vergence mismatches (AVM) at the fixation and peripheral points, which can result in severe visual discomfort. With the aim of alleviating discomfort arising from AVM, we propose a new visual comfort enhancement approach for processing S3D visual signals to deliver a more comfortable 3D viewing experience at the display. This is accomplished via an optimization process whereby a predictive indicator of visual discomfort is minimized, while still aiming to maintain the viewer's sense of 3D presence by performing a suitable parallax shift, and by directed blurring of the signal. Our processing framework is defined on 3D visual coordinates that reflect the nonuniform resolution of retinal sensors and that uses a measure of 3D saliency strength. An appropriate level of blur that corresponds to the degree of parallax shift is found, making it possible to produce synthetic accommodation cues implemented using a perceptively relevant filter. By this method, AVM, the primary contributor to the discomfort felt when viewing S3D images, is reduced. We show via a series of subjective experiments that the proposed approach improves visual comfort while preserving the sense of 3D presence.

Since ultra-high-definition (UHD) display has larger resolution and various display size, it is necessary to measure image sharpness considering variation in visual resolution caused by diverse viewing geometry. In this paper, we propose a no-reference perceptual sharpness assessment model of UHD images. The proposed model analyzes viewing geometry in terms of display resolution and viewing environment. Then, we measure the local adaptive sharpness score in accordance with the textural motion blur, texture, and edge. In addition, we propose a spatial pooling method associated with foveal regions, which is caused by nonuniform distribution of the photoreceptors on a human retina. Through the rigorous experiments, we demonstrate that the proposed model can measure the sharpness of UHD images more accurately than other image sharpness assessment methods.

To maximize the presence experienced by humans, visual content has evolved to achieve a higher visual presence in a series of high definition (HD), ultra HD (UHD), 8K UHD, and 8K stereoscopic 3D (S3D). Several studies have introduced visual presence delivered from content when viewing UHD S3D from a content analysis perspective. Nevertheless, no clear definition has been presented for visual presence, and only a subjective evaluation has been relied upon. The main reason for this is that there is a limitation to defining visual presence via the use of content information itself. In this paper, we define the visual presence for each viewing environment, and investigate a novel methodology to measure the experienced visual presence when viewing both 2D and 3D via the definition of a new metric termed volume of visual information by quantifying the influence of the viewing geometry between the display and viewer. To achieve this goal, the viewing geometry and display parameters for both flat and atypical displays are analyzed in terms of human perception by introducing a novel concept of pixel-wise geometry. In addition, perceptual weighting through analysis of content information is performed in accordance with monocular and binocular vision characteristics. In the experimental results, it is shown that the constructed model based on the viewing geometry, content, and perceptual characteristics has a high correlation of about 84% with subjective evaluations.

The human visual system perceives 3D depth following sensing via its binocular optical system, a series of massively parallel processing units, and a feedback system that controls the mechanical dynamics of eye movements and the crystalline lens. The process of accommodation (focusing of the crystalline lens) and binocular vergence is controlled simultaneously and symbiotically via cross-coupled communication between the two critical depth computation modalities. The output responses of these two subsystems, which are induced by oculomotor control, are used in the computation of a clear and stable cyclopean 3D image from the input stimuli. These subsystems operate in smooth synchronicity when one is viewing the natural world; however, conflicting responses can occur when viewing stereoscopic 3D (S3D) content on fixed displays, causing physiological discomfort. If such occurrences could be predicted, then they might also be avoided (by modifying the acquisition process) or ameliorated (by changing the relative scene depth). Toward this end, we have developed a dynamic accommodation and vergence interaction (DAVI) model that successfully predicts visual discomfort on S3D images. The DAVI model is based on the phasic and reflex responses of the fast fusional vergence mechanism. Quantitative models of accommodation and vergence mismatches are used to conduct visual discomfort prediction. Other 3D perceptual elements are included in the proposed method, including sharpness limits imposed by the depth of focus and fusion limits implied by Panum's fusional area. The DAVI predictor is created by training a support vector machine on features derived from the proposed model and on recorded subjective assessment results. The experimental results are shown to produce accurate predictions of experienced visual discomfort.

Visual discomfort assessment (VDA) on stereoscopic images is of fundamental importance for making decisions regarding visual fatigue caused by unnatural binocular alignment. Nevertheless, no solid framework exists to quantify this discomfort using models of the responses of visual neurons. Binocular vision is realized by means of neural mechanisms that subserve the sensorimotor control of eye movements. We propose a neuronal model-based framework called Neural 3D Visual Discomfort Predictor (N3D-VDP) that automatically predicts the level of visual discomfort experienced when viewing stereoscopic 3D (S3D) images. The N3D-VDP model extracts features derived by estimating the neural activity associated with the processing of binocular disparities. In this regard we deploy a model of disparity processing in the extra-striate middle temporal (MT) region of occipital lobe. We compare the performance of N3D-VDP with other recent VDA algorithms using correlations against reported subjective visual discomfort, and show that N3D-VDP is statistically superior to the other methods.

Being able to predict the degree of visual discomfort that is felt when viewing stereoscopic 3D (S3D) images is an important goal toward ameliorating causative factors, such as excessive horizontal disparity, misalignments or mismatches between the left and right views of stereo pairs, or conflicts between different depth cues. Ideally, such a model should account for such factors as capture and viewing geometries, the distribution of disparities, and the responses of visual neurons. When viewing modern 3D displays, visual discomfort is caused primarily by changes in binocular vergence while accommodation in held fixed at the viewing distance to a flat 3D screen. This results in unnatural mismatches between ocular fixations and ocular focus that does not occur in normal direct 3D viewing. This accommodation vergence conflict can cause adverse effects, such as headaches, fatigue, eye strain, and reduced visual ability. Binocular vision is ultimately realized by means of neural mechanisms that subserve the sensorimotor control of eye movements. Realizing that the neuronal responses are directly implicated in both the control and experience of 3D perception, we have developed a model-based neuronal and statistical framework called the 3D visual discomfort predictor (3D-VDP)that automatically predicts the level of visual discomfort that is experienced when viewing S3D images. 3D-VDP extracts two types of features: 1) coarse features derived from the statistics of binocular disparities and 2) fine features derived by estimating the neural activity associated with the processing of horizontal disparities. In particular, we deploy a model of horizontal disparity processing in the extrastriate middle temporal region of occipital lobe. We compare the performance of 3D-VDP with other recent discomfort prediction algorithms with respect to correlation against recorded subjective visual discomfort scores,and show that 3D-VDP is statistically superior to the other methods.

Compressive sensing (CS) makes it possible to more naturally create compact representations of data with respect to a desired data rate. Through wavelet decomposition, smooth and piecewise smooth signals can be represented as sparse and compressible coefficients. These coefficients can then be effectively compressed via the CS. Since a wavelet transform divides image information into layered blockwise wavelet coefficients over spatial and frequency domains, visual improvement can be attained by an appropriate perceptually weighted CS scheme. We introduce such a method in this paper and compare it with the conventional CS. The resulting visual CS model is shown to deliver improved visual reconstructions.

Compressive Sensing (CS) aims to recover a sparse signal from a small number of projections onto random vectors. Because of its great practical possibility, both academia and industries have made efforts to develop the CS's reconstruction performance, but most of existing works remain at the theoretical study. In this paper, we propose a new Block Compres-sive Sensing (nBCS), which has several benefits compared to the general CS methods. In particular, the nBCS can be dynamically adaptive to varying channel capacity because it conveys the good inheritance of the wavelet transform.