Motion Stream Research Articles

Two-Stream Collaborative Learning With Spatial-Temporal Attention for Video Classification

Video classification is highly important and has widespread applications, such as video search and intelligent surveillance. Video naturally contains both static and motion information, which can be represented by frames and optical flow, respectively. Recently, researchers have generally adopted deep networks to capture the static and motion information separately , which has two main limitations. First, the coexistence relationship between spatial and temporal attention is ignored, although they should be jointly modeled as the spatial and temporal evolutions of video to learn discriminative video features. Second, the strong complementarity between static and motion information is ignored, although they should be collaboratively learned to enhance each other. To address the above two limitations, this paper proposes the two-stream collaborative learning with spatial-temporal attention (TCLSTA) approach, which consists of two models. First, for the spatial-temporal attention model , the spatial-level attention emphasizes the salient regions in a frame, and the temporal-level attention exploits the discriminative frames in a video. They are mutually enhanced to jointly learn the discriminative static and motion features for better classification performance. Second, for the static-motion collaborative model , it not only achieves mutual guidance between static and motion information to enhance the feature learning but also adaptively learns the fusion weights of static and motion streams, thus exploiting the strong complementarity between static and motion information to improve video classification. Experiments on four widely used data sets show that our TCLSTA approach achieves the best performance compared with more than 10 state-of-the-art methods.

Spatiotemporal Relation Networks for Video Action Recognition

Two-stream convolutional networks have shown strong performance in a video action recognition task for its ability to capture spatial and temporal features simultaneously. However, the calculation of optical flow is time-consuming and it cannot be applied to the real-time processing of video. To address this problem, this paper proposes a new end-to-end architecture called SpatioTemporal Relation Networks (STRN) to extract spatial information and temporal information simultaneously from the video with the only RGB input. STRN consist of two branches, called appearance stream and motion stream, respectively. Appearance stream retains the structure of the original spatial stream in the two-stream architecture with the input of consecutive frames instead of a single frame. Motion stream, which takes relation information between the adjacent features in the appearance stream as an input, can effectively complement appearance stream. A relation block is an extractor which is used to extract relation information from the appearance stream. STRN can learn spatiotemporal information from the video with the only RGB input, which avoids the calculation of optical flow. We validate the STRN on UCF-101 and HMDB-51 and achieve better performance.

Learning to Segment Moving Objects

We study the problem of segmenting moving objects in unconstrained videos. Given a video, the task is to segment all the objects that exhibit independent motion in at least one frame. We formulate this as a learning problem and design our framework with three cues: (i) independent object motion between a pair of frames, which complements object recognition, (ii) object appearance, which helps to correct errors in motion estimation, and (iii) temporal consistency, which imposes additional constraints on the segmentation. The framework is a two-stream neural network with an explicit memory module. The two streams encode appearance and motion cues in a video sequence respectively , while the memory module captures the evolution of objects over time, exploiting the temporal consistency. The motion stream is a convolutional neural network trained on synthetic videos to segment independently moving objects in the optical flow field. The module to build a 'visual memory' in video, i.e., a joint representation of all the video frames, is realized with a convolutional recurrent unit learned from a small number of training video sequences. For every pixel in a frame of a test video, our approach assigns an object or background label based on the learned spatio-temporal features as well as the 'visual memory' specific to the video. We evaluate our method extensively on three benchmarks, DAVIS, Freiburg-Berkeley motion seg-mentation dataset and SegTrack. In addition, we provide an extensive ablation study to investigate both the choice of the training data and the influence of each component in the proposed framework.

Pixel Objectness: Learning to Segment Generic Objects Automatically in Images and Videos.

We propose an end-to-end learning framework for segmenting generic objects in both images and videos. Given a novel image or video, our approach produces a pixel-level mask for all "object-like" regions-even for object categories never seen during training. We formulate the task as a structured prediction problem of assigning an object/background label to each pixel, implemented using a deep fully convolutional network. When applied to a video, our model further incorporates a motion stream, and the network learns to combine both appearance and motion and attempts to extract all prominent objects whether they are moving or not. Beyond the core model, a second contribution of our approach is how it leverages varying strengths of training annotations. Pixel-level annotations are quite difficult to obtain, yet crucial for training a deep network approach for segmentation. Thus we propose ways to exploit weakly labeled data for learning dense foreground segmentation. For images, we show the value in mixing object category examples with image-level labels together with relatively few images with boundary-level annotations. For video, we show how to bootstrap weakly annotated videos together with the network trained for image segmentation. Through experiments on multiple challenging image and video segmentation benchmarks, our method offers consistently strong results and improves the state-of-the-art for fully automatic segmentation of generic (unseen) objects. In addition, we demonstrate how our approach benefits image retrieval and image retargeting, both of which flourish when given our high-quality foreground maps. Code, models, and videos are at: http://vision.cs.utexas.edu/projects/pixelobjectness/.

Multi-stream CNN: Learning representations based on human-related regions for action recognition

The most successful video-based human action recognition methods rely on feature representations extracted using Convolutional Neural Networks (CNNs). Inspired by the two-stream network (TS-Net), we propose a multi-stream Convolutional Neural Network (CNN) architecture to recognize human actions. We additionally consider human-related regions that contain the most informative features. First, by improving foreground detection, the region of interest corresponding to the appearance and the motion of an actor can be detected robustly under realistic circumstances. Based on the entire detected human body, we construct one appearance and one motion stream. In addition, we select a secondary region that contains the major moving part of an actor based on motion saliency. By combining the traditional streams with the novel human-related streams, we introduce a human-related multi-stream CNN (HR-MSCNN) architecture that encodes appearance, motion, and the captured tubes of the human-related regions. Comparative evaluation on the JHMDB, HMDB51, UCF Sports and UCF101 datasets demonstrates that the streams contain features that complement each other. The proposed multi-stream architecture achieves state-of-the-art results on these four datasets.

Project of forming «culture and safety» of the airport

Using the method analysis of the systems and synthesis parameters that influence on the process evacuation of people were certain. Methodological principles are used in relation to the management of people streams on the basis of the use critical path method, where the optimization synthesis of evacuation route people is considered as a topological chart of technological line, and an evacuation way is broken on separate parts. By means of probabilistic method and optimization synthesis of flexible technological lines topological case of users of airport streams frames are worked out and sometimes the evacuation. Conducted analysis of existent mathematical models, that describe motion streams of people on the objects of their mass stay showed that the stream users of the airport mutated and him mathematical description requires the synthesis of a few mathematical models. The method calculation time of evacuation users of the airport in that person informative base drawn on results other undertaken experimental studies is accordingly offered.

Different trajectories of decline for global form and global motion processing in aging, mild cognitive impairment and Alzheimer's disease

The visual processing of complex motion is impaired in Alzheimer's disease (AD). However, it is unclear whether these impairments are biased toward the motion stream or part of a general disruption of global visual processing, given some reports of impaired static form processing in AD. Here, for the first time, we directly compared the relative preservation of motion and form systems in AD, mild cognitive impairment, and healthy aging, by measuring coherence thresholds for well-established global rotational motion and static form stimuli known to be of equivalent complexity. Our data confirm a marked motion-processing deficit specific to some AD patients, and greater than any form-processing deficit for this group. In parallel, we identified a more gradual decline in static form recognition, with thresholds raised in mild cognitive impairment patients and slightly further in the AD group compared with controls. We conclude that complex motion processing is more vulnerable to decline in dementia than complex form processing, perhaps owing to greater reliance on long-range neural connections heavily targeted by AD pathology.

Functional organization of telencephalic visual association fields in pigeons

Birds show remarkable visual abilities that surpass most of our visual psychophysiological abilities. In this study, we investigated visual associative areas of the tectofugal visual system in pigeons. Similar to the condition in mammals, ascending visual pathways in birds are subdivided into parallel form/color vs. motion streams at the thalamic and primary telencephalic level. However, we know practically nothing about the functional organization of those telencephalic areas that receive input from the primary visual telencephalic fields. The current study therefore had two objectives: first, to reveal whether these visual associative areas of the tectofugal system are activated during visual discrimination tasks; second, to test whether separated form/color vs. motion pathways can be discerned among these association fields. To this end, we trained pigeons to discriminate either form/color or motion stimuli and used the immediate early gene protein ZENK to capture the activity of the visual associative areas during the task. We could indeed identify several visual associative telencephalic structures by activity pattern changes during discriminations. However, none of these areas displayed a difference between form/color vs. motion sessions. The presence of such a distinction in thalamo-telencephalic, but not in further downstream visual association areas opens the possibility that these separate streams converge very early in birds, which possibly minimizes long-range connections due to the evolutionary pressure toward miniaturized brains.

Embodied learning of a generative neural model for biological motion perception and inference

Although an action observation network and mirror neurons for understanding the actions and intentions of others have been under deep, interdisciplinary consideration over recent years, it remains largely unknown how the brain manages to map visually perceived biological motion of others onto its own motor system. This paper shows how such a mapping may be established, even if the biologically motion is visually perceived from a new vantage point. We introduce a learning artificial neural network model and evaluate it on full body motion tracking recordings. The model implements an embodied, predictive inference approach. It first learns to correlate and segment multimodal sensory streams of own bodily motion. In doing so, it becomes able to anticipate motion progression, to complete missing modal information, and to self-generate learned motion sequences. When biological motion of another person is observed, this self-knowledge is utilized to recognize similar motion patterns and predict their progress. Due to the relative encodings, the model shows strong robustness in recognition despite observing rather large varieties of body morphology and posture dynamics. By additionally equipping the model with the capability to rotate its visual frame of reference, it is able to deduce the visual perspective onto the observed person, establishing full consistency to the embodied self-motion encodings by means of active inference. In further support of its neuro-cognitive plausibility, we also model typical bistable perceptions when crucial depth information is missing. In sum, the introduced neural model proposes a solution to the problem of how the human brain may establish correspondence between observed bodily motion and its own motor system, thus offering a mechanism that supports the development of mirror neurons.

Neural substrates underlying the passive observation and active control of translational egomotion.

Moving or static obstacles often get in the way while walking in daily life. Avoiding obstacles involves both perceptual processing of motion information and controlling appropriate defensive movements. Several higher-level motion areas, including the ventral intraparietal area (VIP), medial superior temporal area, parieto-insular vestibular cortex (PIVC), areas V6 and V6A, and cingulate sulcus visual area, have been identified in humans by passive viewing of optic flow patterns that simulate egomotion and object motion. However, the roles of these areas in the active control of egomotion in the real world remain unclear. Here, we used functional magnetic resonance imaging (fMRI) to map the neural substrates underlying the passive observation and active control of translational egomotion in humans. A wide-field virtual reality environment simulated a daily scenario where doors randomly swing outward while walking in a hallway. The stimuli of door-dodging events were essentially the same in two event-related fMRI experiments, which compared passive and active dodges in response to swinging doors. Passive dodges were controlled by a computer program, while active dodges were controlled by the subject. Passive dodges activated several higher-level areas distributed across three dorsal motion streams in the temporal, parietal, and cingulate cortex. Active dodges most strongly activated the temporal-vestibular stream, with peak activation located in the right PIVC. Other higher-level motion areas including VIP showed weaker to no activation in active dodges. These results suggest that PIVC plays an active role in sensing and guiding translational egomotion that moves an observer aside from impending obstacles.

A hand gesture recognition technique for human–computer interaction

We propose an approach to recognize trajectory-based dynamic hand gestures in real time for human–computer interaction (HCI). We also introduce a fast learning mechanism that does not require extensive training data to teach gestures to the system. We use a six-degrees-of-freedom position tracker to collect trajectory data and represent gestures as an ordered sequence of directional movements in 2D. In the learning phase, sample gesture data is filtered and processed to create gesture recognizers, which are basically finite-state machine sequence recognizers. We achieve online gesture recognition by these recognizers without needing to specify gesture start and end positions. The results of the conducted user study show that the proposed method is very promising in terms of gesture detection and recognition performance (73% accuracy) in a stream of motion. Additionally, the assessment of the user attitude survey denotes that the gestural interface is very useful and satisfactory. One of the novel parts of the proposed approach is that it gives users the freedom to create gesture commands according to their preferences for selected tasks. Thus, the presented gesture recognition approach makes the HCI process more intuitive and user specific.

Visual change detection recruits auditory cortices in early deafness

Although cross-modal recruitment of early sensory areas in deafness and blindness is well established, the constraints and limits of these plastic changes remain to be understood. In the case of human deafness, for instance, it is known that visual, tactile or visuo-tactile stimuli can elicit a response within the auditory cortices. Nonetheless, both the timing of these evoked responses and the functional contribution of cross-modally recruited areas remain to be ascertained. In the present study, we examined to what extent auditory cortices of deaf humans participate in high-order visual processes, such as visual change detection. By measuring visual ERPs, in particular the visual MisMatch Negativity (vMMN), and performing source localization, we show that individuals with early deafness (N=12) recruit the auditory cortices when a change in motion direction during shape deformation occurs in a continuous visual motion stream. Remarkably this “auditory” response for visual events emerged with the same timing as the visual MMN in hearing controls (N=12), between 150 and 300ms after the visual change. Furthermore, the recruitment of auditory cortices for visual change detection in early deaf was paired with a reduction of response within the visual system, indicating a shift from visual to auditory cortices of part of the computational process. The present study suggests that the deafened auditory cortices participate at extracting and storing the visual information and at comparing on-line the upcoming visual events, thus indicating that cross-modally recruited auditory cortices can reach this level of computation.

The ‘Social Present’ in Interacting Dyads

The therapeutic relationship is an important change factor in psychotherapy, and disordered communication is a core problem in many patients’ everyday functioning. Our research group has worked on measures of nonverbal behavior that can be used to evaluate social interaction. We performed a number of projects on this embodiment background, addressing nonverbal behavior in dyadic interaction processes in psychotherapy (104 therapy sessions of 70 patients) as well as healthy dyads discussing topics of general interest (two projects included 84 and 51 dyads, respectively). Nonverbal synchrony was assessed objectively using an automated video-analysis algorithm (Motion Energy Analysis, MEA) developed in our laboratory. All participants were unaware of the synchrony measure being assessed. In the initial psychotherapy project, we found the quality of interaction was embodied in the degree of nonverbal synchrony between therapist and patient. In healthy dyads, synchrony was related to the type of interaction and predicted the affective pleasantness of the conversations. This suggested that beyond the mere amount of movement, the degree of nonverbal synchrony between people was a pivotal predictor of features of the interaction, as well as an objective and sensitive indicator of the severity of patients’ problems. In general, social synchronization is an important, usually unattended, capacity that regulates social interaction and expresses the satisfaction with social exchange. Its analysis provides valuable insights into embodied social cognition and produces promising targets for psychotherapeutic and other interventions. Here we report on a recent elaboration based on MEA. This concerned our definition of a duration measure of the social present (’nowness’) of a dyad. The social present was defined as the size of the temporal window within which the nonverbal motion streams of interactands are significantly correlated. A pilot study showed that this duration has an extension of around 5 s. We discuss the associations of the social present with trait and state variables of the participants.

Spatial references and audio-tactile interaction in cross-modal dynamic capture.

In audiotactile dynamic capture, judgment of the direction of an apparent motion stream (such as auditory motion) was impeded (hence 'captured') by the presentation of a concurrent, but directionally opposite apparent motion stream (such as tactile motion) from a distractor modality, leading to a cross-modal dynamic capture (CDC) effect. That is to say, the percentage of correct reporting of the direction of the target motion was reduced. Previous studies have revealed the effect of stimulus onset asynchronies (SOAs) and the potential spatial remapping (by adopting a cross-hands posture) in CDC. However, further exploration of the dynamic capture process under different postures was not available due to the fact that only two levels of time asynchronies were employed (either synchronous or with an SOA of 500 ms). This study introduced a broad range of SOAs (-400 ms to 400 ms, tactile stream preceded auditory stream or vice versa) to explore the time course of audio-tactile interaction in CDC with two spatial references--arms-uncrossed or arms-crossed postures. Participants judged the direction of auditory apparent motion with tactile distractors. The results showed that in the arms-uncrossed condition, the CDC effect was prominent when the auditory-tactile events were in the temporal integration window (0-60 ms). However, with a preceding tactile cueing effect of SOA equal to and above 150 ms, the CDC effect was reduced, and no CDC effect was observed with the arms-crossed posture. These results suggest CDC effect is modulated by both cross-modal interaction and the spatial reference (especially for the distractors). The magnitude of the CDC effects in audiotactile interaction may be accounted for by reliability of tactile spatial-temporal information.

Object motion continuity and the flash-lag effect

When a visual object is briefly flashed, it appears to lag behind another moving object (flash-lag effect; FLE). Previous studies showed that a sudden change to the moving object at the time of the flash presentation would eliminate the FLE. We examined whether the FLE would be eliminated when a sudden color change was embedded in a sequence of color alternations on a moving object. Observers viewed a moving disc, the color of which did not change at all, changed only once when another stationary object flashed, or alternated regularly (Experiment 1) or randomly (Experiment 2) between two colors as it was moving before the flash presentation. Although the magnitude was reduced compared with the no-change condition, the FLE observed with the moving object that changed color during motion was significantly stronger than that in the one-change condition. In Experiment 3, the object color alternated between two but unexpectedly changed to a new color when the flash appeared. The elimination of FLE in such condition was comparable to the one-change condition, suggesting that the unexpected change restored the salience of the moving object at the time of flash presentation, which spared the observer from perceiving the FLE. We propose that, without an unexpected event, rapid changes in the surface feature of the moving object partially degrade the maintenance of object file, but this does not preclude the visual system in registering the existence of only one object in the motion stream.

SU-E-U-08: Presentation of a New Intrafractional Prostate Monitoring Method with Ultrasound Image Guidance During Radiotherapy Treatment

Purpose: Clinical presentation of a new intrafractional ultrasound image guidance (USIG) prostate monitoring/tracking method. Methods: Clarity ultrasound system has recently released a new feature to monitor/track intrafractional prostate motion using a trans-perineal image acquisition position. After initial localization and image guidance correction, the ultrasound probe remains fixed in the sagittal-plane, trans-perineal imaging position via a couch-mounted arm-support system. The software then enters a live monitoring mode, where the scanning ultrasound probe continuously acquires fanned sagittal images for tracking of during-treatment prostate position. Compared to other tracking technologies (e.g. RF implanted beacon tracking, or real-time fluoro monitoring), the ultrasound system has the unique advantages of live 3D image display, without invasive procedure or imaging radiation dose. We present our initial experience using this monitoring feature for 8 patients and 218 treatment sessions. Results: The monitoring software functioned as expected during treatment with consistent reporting of prostate deviation from isocenter location, and notification of instances where positional error exceeded our tolerance of 3 mm for 5 second. Comparisons of prostate tracking data streams obtained from Clarity USIG with data streams obtained in our clinic from Calypso RF tracking of a similar patient population show similarity of recorded motion information for both methods. Average during-treatment prostate motion for both tracking methods was on the order of 1–2 mm in 3 cardinal directions, with roughly 5% of instances where motion exceeded 3 mm, 5 second tolerance level. When Monitoring reported that tolerance was exceeded we confirmed this by performing redundant, static imaging Alignment, with confirmation that target position had, indeed, changed as reported. Conclusion: The newly released trans-perineal ultrasound Monitoring approach was confirmed to function well clinically, and to report accurate intrafractional monitoring data. Acquired motion streams were consistent with our Calypso monitoring experience obtained on a similar patient population in our own clinic.

Mutual Influences of Intermodal Visual/Tactile Apparent Motion and Auditory Motion with Uncrossed and Crossed Arms

Intra-modal apparent motion has been shown to be affected or 'captured' by information from another, task-irrelevant modality, as shown in cross-modal dynamic capture effect. Here we created inter-modal apparent motion between visual and tactile stimuli and investigated whether there are mutual influences between auditory apparent motion and inter-modal visual/tactile apparent motion. Moreover, we examined whether and how the spatial remapping between somatotopic and external reference frames of tactile events affect the cross-modal capture between auditory apparent motion and inter-modal visual/tactile apparent motion, by introducing two arm postures: arms-uncrossed and arms-crossed. In Experiment 1, we used auditory stimuli (auditory apparent motion) as distractors and inter-modal visual/tactile stimuli (inter-modal apparent motion) as targets while in Experiment 2 we reversed the distractors and targets. In Experiment 1, we found a general detrimental influence of arms-crossed posture in the task of discrimination of direction in visual/tactile stream, but in Experiment 2, the influence of arms-uncrossed posture played a significant role in modulating the inter-modal visual/tactile stimuli capturing over auditory apparent motion. In both Experiments, the synchronously presented motion streams led to noticeable directional congruency effect in judging the target motion. Among the different modality combinations, tactile to tactile apparent motion (TT) and visual to visual apparent motion (VV) are two signatures revealing the asymmetric congruency effects. When the auditory stimuli were targets, the congruency effect was largest with VV distractors, lowest with TT distractors; the pattern was reversed when the auditory stimuli were distractors. In addition, across both experiments the congruency effect in visual to tactile (VT) and tactile to visual (TV) apparent motion was intermediate between the effect-sizes in VV and TT. We replicated the above findings with a block-wise design (Experiment 3). In Experiment 4, we introduced static distractor events (visual or tactile stimulus), and found the modulation of spatial remapping of distractors upon AA motion is reduced. These findings suggest that there are mutual but a robust asymmetric influence between intra-modal auditory apparent motion and intermodal visual/tactile apparent motion. We proposed that relative reliabilities in directional information between distractor and target streams, summed over a remapping process between two spatial reference frames, determined this asymmetric influence.

Investigating the In-Between: Multisensory Integration of Auditory and Visual Motion Streams

We investigated audiovisual interactions in motion perception by behavioral experiments testing both, the influence of visual stimuli on auditory apparent motion and the influence of auditory stimuli on visual apparent motion perception. A set of loudspeakers with an LED mounted in the middle of each speaker cone was arranged in a semicircle. Apparent motion streams were presented for each modality alone in the unimodal conditions. In the bimodal conditions, stimuli of the second modality were added to fill the temporal and spatial gaps of the sampled trajectory of the reference stream. The participants' task was to observe the quasi-naturalistic stimulus sequences and to perform a standard classification. The addition of stimuli of the second modality indeed facilitated apparent motion perception. Bimodal presentation increased the upper temporal interval up to which the stimuli could be separated in time while still being perceived as continuous motion. We interpret these results as evidence for an ecologically advantageous audiovisual motion integration mechanism which operates beyond the constraints of strict spatiotemporal coincidence. Functional considerations suggest that this mechanism may represent an amodal stage suited for the processing of both unimodal and bimodal signals.

ERP evidence for crossmodal interactions during the encoding of audio–visual motion offsets

Previous behavioural studies have challenged the notion of unrestricted visual dominance in motion perception by demonstrating that the auditory modality can affect visual motion perception in the peripheral field. Thus far, electrophysiological evidence for the interplay between both modalities across space has not been provided, yet. The present study investigated crossmodal interactions during the encoding of bimodal motion offsets at different locations in space. To this end, moving audio–visual stimuli with either congruent or spatially and temporally disparate motion offsets were presented along different azimuthal trajectories in free-field while event-related potentials (ERPs) were recorded. The pattern of interactions at motion offset suggests that dominance effects between modalities are reflected in a latency shift of ERP components. The latencies of the visual offset N1 were shifted towards the respective termination of the moving acoustic signal at peripheral motion offset locations. The presence of a concurrent visual motion stream less consistently affected the latencies of auditory offset components at frontal motion offset locations which only partially supports the widely reported visual dominance in the central visual field. However, the results provide electrophysiological evidence that the auditory system can modulate the processing of motion streams in the peripheral visual field which is supposed to be due to superior tracking of an ongoing motion in the periphery by the auditory system compared to the visual system. Our findings are well in line with previous studies that highlighted the crucial role of the auditory modality when the reliability of the visual signal is degraded.

Non-retinotopic perceptual learning

Perceptual learning is usually specific for the encoded stimuli. For example, improvements in performance for a certain stimulus do not transfer if the stimulus is rotated by 90 degree or is presented at a different location. These findings are usually taken as evidence that orientation-specific and retinotopic synaptic changes underlie perceptual learning. To test this hypothesis, we used a recently developed non-retinotopic masking paradigm. An offset vernier was followed by nine consecutive pairs of non-offset flanking lines arranged along a circular trajectory, giving rise to the percept of two diverging and turning motion streams. Critically, the vernier itself was invisible but its offset was perceived at the last line of the stream, which had a different orientation and location than the vernier. For 4,000 training trials, sixteen observers attended to one motion stream and indicated the perceived offset direction (left vs. right). Performance improved significantly. In addition, we measured performance for various other stimulus configurations before and after training. Our results show that the improvement in performance is specific for the orientation and location of the perceived offset of the last line of the attended motion stream but not for the actual orientation and location of the invisible vernier. Hence, orientation-specific, retinotopic synaptic changes cannot explain our results. We propose instead that perceptual learning involves changes in non-retinotopic, attentional readout processes.