◾ Symmetry, Shape, Surfaces, and Objects

Discerning objects from their surrounds (i.e., figure-ground segmentation) in a way that guides adaptive behaviors is a fundamental task of the brain. Neurophysiological work has revealed a class of cells in the macaque visual cortex that may be ideally suited to support this neural computation: border ownership cells (Zhou H, Friedman HS, von der Heydt R. J Neurosci 20: 6594–6611, 2000). These orientation-tuned cells appear to respond conditionally to the borders of objects. A behavioral correlate supporting the existence of these cells in humans was demonstrated with two-dimensional luminance-defined objects (von der Heydt R, Macuda T, Qiu FT. J Opt Soc Am A Opt Image Sci Vis 22: 2222–2229, 2005). However, objects in our natural visual environments are often signaled by complex cues, such as motion and binocular disparity. Thus for border ownership systems to effectively support figure-ground segmentation and object depth ordering, they must have access to information from multiple depth cues with strict depth order selectivity. Here we measured in humans (of both sexes) border ownership-dependent tilt aftereffects after adaptation to figures defined by either motion parallax or binocular disparity. We find that both depth cues produce a tilt aftereffect that is selective for figure-ground depth order. Furthermore, we find that the effects of adaptation are transferable between cues, suggesting that these systems may combine depth cues to reduce uncertainty (Bülthoff HH, Mallot HA. J Opt Soc Am A 5: 1749–1758, 1988). These results suggest that border ownership mechanisms have strict depth order selectivity and access to multiple depth cues that are jointly encoded, providing compelling psychophysical support for their role in figure-ground segmentation in natural visual environments.NEW & NOTEWORTHY Figure-ground segmentation is a critical function that may be supported by “border ownership” neural systems that conditionally respond to object borders. We measured border ownership-dependent tilt aftereffects to figures defined by motion parallax or binocular disparity and found aftereffects for both cues. These effects were transferable between cues but selective for figure-ground depth order, suggesting that the neural systems supporting figure-ground segmentation have strict depth order selectivity and access to multiple depth cues that are jointly encoded.

Binocular disparity is the main depth cue that makes stereoscopic images appear 3D. However, in many scenarios, the range of depth that can be reproduced by this cue is greatly limited and typically fixed due to constraints imposed by displays. For example, due to the low angular resolution of current automultiscopic screens, they can only reproduce a shallow depth range. In this work, we study the motion parallax cue, which is a relatively strong depth cue, and can be freely reproduced even on a 2D screen without any limits. We exploit the fact that in many practical scenarios, motion parallax provides sufficiently strong depth information that the presence of binocular depth cues can be reduced through aggressive disparity compression. To assess the strength of the effect we conduct psycho-visual experiments that measure the influence of motion parallax on depth perception and relate it to the depth resulting from binocular disparity. Based on the measurements, we propose a joint disparity-parallax computational model that predicts apparent depth resulting from both cues. We demonstrate how this model can be applied in the context of stereo and multiscopic image processing, and propose new disparity manipulation techniques, which first quantify depth obtained from motion parallax, and then adjust binocular disparity information accordingly. This allows us to manipulate the disparity signal according to the strength of motion parallax to improve the overall depth reproduction. This technique is validated in additional experiments.

Over the past few years, Generative Adversarial Networks (GANs) have garnered increased interest among researchers in Computer Vision, with applications including, but not limited to, image generation, translation, imputation, and super-resolution. Nevertheless, no GAN-based method has been proposed in the literature that can successfully represent, generate or translate 3D facial shapes (meshes). This can be primarily attributed to two facts, namely that (a) publicly available 3D face databases are scarce as well as limited in terms of sample size and variability (e.g., few subjects, little diversity in race and gender), and (b) mesh convolutions for deep networks present several challenges that are not entirely tackled in the literature, leading to operator approximations and model instability, often failing to preserve high-frequency components of the distribution. As a result, linear methods such as Principal Component Analysis (PCA) have been mainly utilized towards 3D shape analysis, despite being unable to capture non-linearities and high frequency details of the 3D face—such as eyelid and lip variations. In this work, we present 3DFaceGAN, the first GAN tailored towards modeling the distribution of 3D facial surfaces, while retaining the high frequency details of 3D face shapes. We conduct an extensive series of both qualitative and quantitative experiments, where the merits of 3DFaceGAN are clearly demonstrated against other, state-of-the-art methods in tasks such as 3D shape representation, generation, and translation.

Recovering 3D shapes of deformable objects from single 2D images is an extremely challenging and ill-posed problem. Most existing approaches are based on structure-from-motion or graph inference, where a 3D shape is solved by fitting 2D keypoints/mask instead of directly using the vital cue in the original 2D image. These methods usually require multiple views of an object instance and rely on accurate labeling, detection, and matching of 2D keypoints/mask across multiple images. To overcome these limitations, we make effort to reconstruct 3D deformable object shapes directly from the given unconstrained 2D images. In training, instead of using multiple images per object instance, our approach relaxes the constraint to use images from the same object category (with one 2D image per object instance). The key is to disentangle the category-specific representation of the 3D shape identity and the instance-specific representation of the 3D shape displacement from the 2D training images. In testing, the 3D shape of an object can be reconstructed from the given image by deforming the 3D shape identity according to the 3D shape displacement. To achieve this goal, we propose a novel convolutional encoder-decoder network—the Disentangling Deep Network (DisDN). To demonstrate the effectiveness of the proposed approach, we implement comprehensive experiments on the challenging PASCAL VOC benchmark and use different 3D shape ground-truth in training and testing to avoiding overfitting. The obtained experimental results show that DisDN outperforms other state-of-the-art and baseline methods.

Visual perception often fails to recover the veridical 3D shape of objects in the environment due to ambiguity and variability in the available depth cues. However, we rely heavily on 3D shape estimates when planning movements, for example reaching to pick up an object from a slanted surface. Given the wide variety of distortions that can affect 3D perception, how do our actions remain accurate across different environments? One hypothesis is that the visuomotor system performs selective filtering of 3D information to minimize distortions. Indeed, some studies have found that actions appear to preferentially target stereo information when it is put in conflict with texture information. However, since these studies analyze averages over multiple trials, this apparent preference could be produced by sensorimotor adaptation. In Experiment 1, we create a set of cue-conflict stimuli where one available depth cue is affected by a constant bias. Sensory feedback rapidly aligns the motor output with physical reality in just a few trials, which can make it seem as if action planning selectively relies on the reinforced cue--yet no change in the relative influences of the cues is necessary to eliminate the constant errors. In contrast, when one depth cue becomes less correlated with physical reality, variable movement errors will occur, causing canonical adaptation to fail as the opposite error corrections cancel out. As a result, canonical adaptation cannot explain the preference for stereo found in studies with variable errors. However, Experiment 2 shows that persistent errors can produce a novel form of adaptation that gradually reduces the relative influence of an unreliable depth cue. These findings show that grasp control processes are continuously modified based on sensory feedback to compensate for both biases and noise in 3D visual processing, rather than having a hardwired preference for one type of depth information.

The ability to detect surfaces was studied in a multiple-cue condition in which binocular disparity and motion parallax could specify independent depth configurations. On trials on which binocular disparity and motion parallax were presented together, either binocular disparity or motion parallax could indicate a surface in one of two intervals; in the other interval, both sources indicated a volume of random points. Surface detection when the two sources of information were present and compatible was not better than detection in baseline conditions, in which only one source of information was present. When binocular disparity and motion specified incompatible depths, observers' ability to detect a surface was severely impaired if motion indicated a surface but binocular disparity did not. Performance was not as severely degraded when binocular disparity indicated a surface and motion did not. This dominance of binocular disparity persisted in the presence of foreknowledge about which source of information would be relevant.

We used functional magnetic resonance imaging to investigate the human cortical areas involved in processing 3-dimensional (3D) shape from texture (SfT) and shading. The stimuli included monocular images of randomly shaped 3D surfaces and a wide variety of 2-dimensional (2D) controls. The results of both passive and active experiments reveal that the extraction of 3D SfT involves the bilateral caudal inferior temporal gyrus (caudal ITG), lateral occipital sulcus (LOS) and several bilateral sites along the intraparietal sulcus. These areas are largely consistent with those involved in the processing of 3D shape from motion and stereo. The experiments also demonstrate, however, that the analysis of 3D shape from shading is primarily restricted to the caudal ITG areas. Additional results from psychophysical experiments reveal that this difference in neuronal substrate cannot be explained by a difference in strength between the 2 cues. These results underscore the importance of the posterior part of the lateral occipital complex for the extraction of visual 3D shape information from all depth cues, and they suggest strongly that the importance of shading is diminished relative to other cues for the analysis of 3D shape in parietal regions.

Interhemispheric depth judgement

In perceiving 3D shape from ambiguous shading patterns, humans use the prior knowledge that the light is located above their head and slightly to the left. Although this observation has fascinated scientists and artists for a long time, the neural basis of this "light from above left" preference for the interpretation of 3D shape remains largely unexplored. Combining behavioral and functional MRI measurements coupled with multivoxel pattern analysis, we show that activations in early visual areas predict best the light source direction irrespective of the perceived shape, but activations in higher occipitotemporal and parietal areas predict better the perceived 3D shape irrespective of the light direction. These findings demonstrate that illumination is processed earlier than the representation of 3D shape in the visual system. In contrast to previous suggestions, we propose that prior knowledge about illumination is processed in a bottom-up manner and influences the interpretation of 3D structure at higher stages of processing.

Independent of edges and 2-D shape that can be highly informative of object identity, depth cues alone can also give rise to vivid and effective object percepts. The processing of different depth cues engages segregated cortical areas, and an efficient object representation would be one that is invariant to depth cues. Here, we investigated depth-cue invariance of object representations by measuring the category-specific response to faces-the M170 response measured with magnetoencephalography. The M170 response is strongest to faces and is sensitive to adaptation, such that repeated presentation of a face diminishes subsequent M170 responses. We used this feature of the M170 and measured the degree to which the adaptation effect is affected by variations in depth cue and 3-D object shape. Subjects viewed a rapid presentation of two stimuli-an adaptor and a test stimulus. The adaptor was either a face, a chair, or a face-like oval surface, and rendered with a single depth cue (shading, structure from motion, or texture). The test stimulus was always a shaded face of a random identity, thus completely controlling for low-level influences on the M170 response to the test stimulus. In the left fusiform face area, we found strong M170 adaptation when the adaptor was a face regardless of its depth cue. This adaptation was marginal in the right fusiform and negligible in the occipital regions. Our results support the presence of depth-cue-invariant representations in the human visual system, alongside size, position, and viewpoint invariance.

Highly discriminative 3D shape representations can be formed by encoding the spatial relationship among virtual words into the Bag of Words (BoW) method. To achieve this challenging task, several unresolved issues in the encoding procedure must be overcome for 3D shapes, including: 1) arbitrary mesh resolution; 2) irregular vertex topology; 3) orientation ambiguity on the 3D surface; and 4) invariance to rigid and non-rigid shape transformations. In this paper, a novel spatially enhanced 3D shape representation called bag of spatial context correlations (BoSCCs) is proposed to address all these issues. Adopting a novel local perspective, BoSCC is able to describe a 3D shape by an occurrence frequency histogram of spatial context correlation patterns, which makes BoSCC become more compact and discriminative than previous global perspective-based methods. Specifically, the spatial context correlation is proposed to simultaneously encode the geometric and spatial information of a 3D local region by the correlation among spatial contexts of vertices in that region, which effectively resolves the aforementioned issues. The spatial context of each vertex is modeled by Markov chains in a multi-scale manner, which thoroughly captures the spatial relationship by the transition probabilities of intra-virtual words and the ones of inter-virtual words. The high discriminability and compactness of BoSCC are effective for classification and retrieval, especially in the scenarios of limited samples and partial shape retrieval. Experimental results show that BoSCC outperforms the state-of-the-art spatially enhanced BoW methods in three common applications: global shape retrieval, shape classification, and partial shape retrieval.

Linear discrimination dictionary learning for shape descriptors

Neural mechanisms of three-dimensional vision

Can movement parallax compensate lacking stereopsis in spatial explorative search tasks?

The ability of younger and older observers to perceive surface slant was investigated in four experiments. The surfaces possessed slants of 20 degrees, 35 degrees, 50 degrees, and 65 degrees, relative to the frontoparallel plane. The observers judged the slants using either a palm board (Experiments 1, 3, and 4) or magnitude estimation (Experiment 2). In Experiments 1-3, physically slanted surfaces were used (the surfaces possessed marble, granite, pebble, and circle textures), whereas computer-generated 3-D surfaces (defined by motion parallax and binocular disparity) were utilized in Experiment 4. The results showed that the younger and older observers' performance was essentially identical with regard to accuracy. The younger and older age groups, however, differed in terms of precision in Experiments 1 and 2: The judgments of the older observers were more variable across repeated trials. When taken as a whole, the results demonstrate that older observers (at least through the age of 83 years) can effectively extract information about slant in depth from optical patterns containing texture, motion parallax, or binocular disparity.