Search

The present study investigated the impact of the dominant hand on performance in two paper-and-pencil tests of visual selective attention (d2-R; FAIR-2). We hypothesized that preview benefits (i.e., preprocessing of stimuli located to the right of the currently fixated item) could improve test performance of left-handers as compared to right-handers because using the right hand could prevent preview benefits simply by covering subsequent stimuli. A group of left-handed students (n = 86) and a group of right-handed students (n = 90) completed both the test d2-R and the test FAIR-2 with their dominant (writing) hand. Results showed that left-handed participants outperformed right-handers in both tests. Hence, when the results of left-handers are to be compared to right-handers or to normative data (which are dominated by right-handers), the impact of preview benefits on left-hander performance should be taken into account.

This study examines the human gloss perception of printing papers at various illuminances and distances from a light source to the object’s surface. Gloss is evaluated based on not only the intensity of reflected light but also the sharpness of specular highlights. The apparent spread of the reflected light source image, which is also used for gloss evaluations, depends on the distance between the light source and the object’s surface. Unlike physical variation of specular image properties, the perception of gloss may exhibit constancy similar to color perception. Our results reveal that illuminance has a strong effect on gloss perception. We found cases where low-gloss samples looked glossier than high-gloss ones—the gloss reversal phenomenon. These results suggest that there is a case in which gloss constancy may not work in every condition.

Biederman and Cooper (1991a) showed that the presentation of a briefly presented image of an object at one position in the visual field facilitated its identification, as assessed by naming speed and accuracy, several minutes later. The facilitation was unaffected by a translation or a reflection of the stimulus. A component of this priming was visual rather than basic-level conceptual or lexical in that there was less facilitation for an object with the same name (and basic-level class) but a different shape. The invariance of priming to view variables has stood up well over the years and appears to be a general phenomenon--as long as the original structural description can be readily resolved--in that it has also been observed for variations in size and orientation in depth. Although priming was unaffected by a change in position, we documented that there was explicit memory for the position (and orientation and size) of the stimulus. The existence of two forms of representation from the identical stimulus presentation--one invariant and the other dependent on view variables--poses a challenge as to what can be concluded about view invariance from single-unit activity.

One function of the central executive component of the working memory model is the coordination of two simultaneous tasks (the dual-task method). The concept of a coordination function has proved effective in identifying brain damaged patients with severe behavioural problems and in discriminating the performance of patients with mild Alzheimer's disease from that of healthy elderly. A computerised version of the dual-task method was successfully used in previous laboratory studies but it is not suited to use in a practical, clinical setting. Thus a paper and pencil version of the dual-task method was devised. The present study was aimed at refining the paper and pencil dual-task method; investigating whether or not age per se affects performance; assessing its test-retest reliability and providing norms for the Georgian population. The new paper and pencil version of the dual-task was administered to 64 young and 64 middle-age and elderly participants. A new, modified, simpler version of the paper and pencil dual-task, the "Tbilisi paper and pencil motor task" was developed. This task has high test-retest reliability and correlates with the established computerised version of the task. No age or education effects were detected.

A new measure of selective attention was presented to 259 normal volunteers. The test was derived from earlier work which emphasized the importance of controlling the distracting conditions using a selective-attention paradigm. In the present study a paper-and-pencil, digit-cancellation format was introduced and the data documented a significant dissociation between tasks analogous to controlled search and automatic detection, two theoretically important aspects of selective attention. Presented is a discussion of similarities and differences of the new measure and classical methods of discriminating between controlled search and automatic detection and the potential significance of this discrimination in clinical neuropsychological investigations.

482 articles and books are presented in alphabetical order. The bibliographaphy surveys the English language literature on early perceptual development in animals and humans during the period 1967-1974, inclusive.

Kok (1986) investigated why P3 amplitudes are reduced when stimuli are degraded. He proposed that a positive moment-related component might overlap P3 after intact stimuli and a negative one after degraded stimuli ("MRP overlap" hypothesis). Discussing his findings, it is suggested that MRP overlap can only account for the result if another overlap hypothesis ("nogo overlap") is added. But even this combined hypothesis cannot explain reducing effects of degradation found in other paradigms. While other overlap hypotheses have been proposed in the literature, the most parsimonious hypothesis is to assume that P3 itself gets smaller after degraded stimuli.

Motor experiences and active exploration during early childhood may affect individual differences in a wide range of perceptual and cognitive abilities. In the current study, we suggest that active exploration of objects facilitates the ability to process object forms and magnitudes, which in turn impacts the development of numerosity perception. We tested our hypothesis by conducting a preregistered active exploration intervention with 59 8‐month‐old infants. The minimal intervention consisted of actively playing with and exploring blocks once a day for 8 weeks. In order to control for possible training effects on attention, we used book reading as a control condition. Pre‐ and post‐test assessments using eye‐tracking showed that block play improved visual form perception, where infants became better at detecting a deviant shape. Furthermore, using three control tasks, we showed that the intervention specifically improved infants' ability to process visual forms and the effect could not be explained by a domain general improvement in attention or visual perception. We found that the intervention did not improve numerosity perception and suggest that because of the sequential nature of our hypothesis, a longer time frame might be needed to see improvements in this ability. Our findings indicate that if infants are given more opportunities for play and exploration, it will have positive effects on their visual form perception, which in turn could help their understanding of geometrical concepts., Playing with blocks leads 8‐month‐old infants to develop better visual form perception, conceptualized here as detecting a deviant shape, compared to infants who were assigned to a reading control group.

Svrha rada: Željela se ustanoviti razlika u percepciji estetike prednjih gornjih zuba s obzirom na anatomske varijacije, oblik zubnih lukova, izgled gingive i linije osmijeha, te općenito izgled donje trećine lica između triju skupina procjenjivača. Materijali i metode: U istraživanje je bilo uključeno 60 ispitanika s trajnom potpunom denticijom, bez protetičkih radova i s Angleovom klasom I, čija je donja trećina lica tijekom osmijeha fotografirana. Te fotografije zatim su procjenjivale osobe iz opće populacije (21), doktori dentalne medicine (20) i specijalisti stomatološke protetike (20) koristeći se modificiranim OES upitnikom (Orofacijalna estetska ljestvica). Rezultati: Procjenjivači iz opće populacije najlošije su ocijenili oblik zuba, boju zuba i izgled zubnih lukova, a najbolje izgled zubnog mesa, a protetičari su pak najbolje ocjene dali (p < 0,01) obliku i boji zuba, izgledu zubnih lukova i usnica te općenito izgledu donje trećine lica, a najlošije izgledu zubnog mesa (p < 0,01). Muškarci i osobe u dobi od 36 do 55 godina najbolje su ocijenili oblik i boju zuba, izgled zubnih lukova i usnica te općenito izgled donje trećine lica. U procjeni zubnog mesa najtolerantniji su bili ocjenjivači u dobi iznad 55 godina (p < 0,01). Zaključak: Uzimajući u obzir dob, spol i profesionalnu izobrazbu, može se zaključiti da svi kriteriji imaju svoj doprinos u procjeni estetike zuba i okolnih struktura. Rezultati ovog istraživanja mogu pomoći pri planiranju stomatoloških zahvata., Purpose: To determine the differences in the perception of aesthetic characteristics of the front teeth which include anatomic variations, appearance of the gums and dental arches, smiles and the lower third of the face between three groups of assessors. Materials and Methods: The study included 60 patients with permanent fully toothed dentition with no prosthetic work done and with Angle class I, whose lower third of the face was photographed while smiling. Collected photos were then assessed by people of the general population (21), graduate doctors of dental medicine (20) and specialists in dental prosthetics (20) using a modified OAS questionnaire (Orofacial aesthetic scale). Results: General population assessors gave weakest scores to the appearance of teeth, tooth color and appearance of the dental arches, and they gave highest scores to the appearance of the gums. Specialists in dental prosthetics gave top scores (p

Recognition of an object usually involves a wide range of sensory inputs. Accumulating evidence shows that first brain responses associated with the visual discrimination of objects emerge around 150 ms, but fewer studies have been devoted to measure the first neural signature of haptic recognition. To investigate the speed of haptic processing, we recorded event-related potentials (ERPs) during a shape discrimination task without visual information. After a restricted exploratory procedure, participants ( n = 27) were instructed to judge whether the touched object corresponded to an expected object whose name had been previously presented in a screen. We encountered that any incongruence between the presented word and the shape of the object evoked a frontocentral negativity starting at ∼175 ms. With the use of source analysis and L2 minimum-norm estimation, the neural sources of this differential activity were located in higher level somatosensory areas and prefrontal regions involved in error monitoring and cognitive control. Our findings reveal that the somatosensory system is able to complete an amount of haptic processing substantial enough to trigger conflict-related responses in medial and prefrontal cortices in

The goal of this review is to start to consolidate and distill the substantial body of research that comprises the published work of the late Professor Steven S. Hsiao. The studies of Hsiao began by demonstrating the receptive field properties of somatosensory neurons, progressed to describing cortical feature selectivity, and then eventually elevated the field to hopes of tapping into natural neural codes with artificial somatosensory feedback. With ongoing analogies to contemporaneous studies in visual neuroscience, the research results and writings of Hsiao have provided the fields of haptics and somatosensory neurophysiology with the conceptual tools needed to allow profound progress. Specifically, Hsiao suggested that slowly adapting tactile form perception could be restored with cortical microstimulation, rapidly adapting slip reflexes should be relegated to low-level, hard-wired prosthetic components, and Pacinian-corpuscle spatiotemporal population responses could potentially be decoded/encoded to provide information about interactions of hands and hand-held instruments with external objects. Future studies will be guided by these insightful reports from Hsiao.

Hippocampal activation is required for episodic memory. Encoding and retrieval of novel and memorable items have been related to different locations in the hippocampus; however, the data remain ambiguous. The application of a newly designed keyboard allowed investigation of brain activation during encoding and free immediate and delayed recall with functional magnetic resonance imaging (fMRI) in young healthy controls (n = 12). Because of the repetitive learning and recall conditions, an individual learning gradient was used to contrast neural activity at different individual levels of novelty. During learning, subjects were asked to memorize 10 geometric patterns requiring the establishment of intra-item associations for memorization. After learning, subjects were asked to recall the items actively via the keyboard. Learning and recall were alternated five times. Delayed recall was scanned about 15 min after the fifth immediate recall condition without subjects having seen the items again. Left-sided anterior hippocampal activity was observed during conditions of initial learning as well as maximum recall. Neural activity during delayed recall did not reveal hippocampal responses and was characterized by a transition of neural activity from occipitoparietal regions to bilateral temporal cortices. We conclude that both lateralization and segregation depend on the specific relational characteristics of the stimuli requiring establishment of intra-item associations for encoding as well as retrieval. The absence of hippocampal activation during delayed recall together with the increase of lateral temporal involvement possibly corresponds with an emerging transition from episodic to long-term memory.

Using a conditioned food avoidance learning paradigm, six squirrel monkeys (Saimiri sciureus) and six common marmosets (Callithrix jacchus) were tested for their ability to (1) reliably form associations between visual or olfactory cues of a potential food and its palatability and (2) remember such associations over prolonged periods of time. We found (1) that at the group level both species showed one-trial learning with the visual cues color and shape, whereas only the marmosets were able to do so with the olfactory cue, (2) that all individuals from both species learned to reliably avoid the unpalatable food items within 10 trials, (3) a tendency in both species for quicker acquisition of the association with the visual cues compared with the olfactory cue, (4) a tendency for quicker acquisition and higher reliability of the aversion by the marmosets compared with the squirrel monkeys, and (5) that all individuals from both species were able to reliably remember the significance of the visual cues, color and shape, even after 4 months, whereas only the marmosets showed retention of the significance of the olfactory cues for up to 4 weeks. Furthermore, the results suggest that in both species tested, illness is not a necessary prerequisite for food avoidance learning but that the presumably innate rejection responses toward highly concentrated but nontoxic bitter and sour tastants are sufficient to induce robust learning and retention.

Several recent studies have documented that non-human primates can individuate objects according to property and/or kind information in much the same way as human infants do from around one year of age when they begin to acquire language. Some studies suggest, however, that only some properties are used for the individuation of food items: color, but not shape. The present study investigated whether these findings reveal a true competence problem with shape properties in the food domain or whether they merely reveal a performance problem (e.g., lack of attention to shapes). We tested 25 great apes (chimpanzees, bonobos and gorillas) in two food individuation tasks. We manipulated subjects’ experience with differences in color and shape properties of food items. Results indicated (i) that all subjects, regardless of their prior experience, solved the color-based object individuation task and (ii) that only the group with previous experience with different shape properties succeeded in the shape-based individuation task. Great apes can thus be primed to take shape into account for individuating food objects, and this results clearly speaks in favor of a performance (rather than a competence) problem in using shape for object individuation of food items. peerReviewed

The crucial role of the medial temporal lobe (MTL) in episodic memory is well established. Although there is little doubt that its anatomical subregions—the hippocampus, peri-, entorhinal and parahippocampal cortex (PHC)—contribute differentially to mnemonic processes, their specific functions in episodic memory are under debate. Data from animal, human lesion, and neuroimaging studies suggest somewhat contradictory perspectives on this functional specialization: a general participation in declarative memory, an exclusive involvement in associative mnemonic processes, and a specific contribution to spatial memory are reported for the hippocampus, adjacent cortices, and the PHC. A functional lateralization in humans dependent on the verbalizability of the material is also discussed herein. To further elucidate the differential contributions of the various MTL subregions to encoding, we employed an object-location association memory paradigm. The memory for each of the studied associations was tested twice: by the object, and by the location serving as retrieval cue. The memory accuracy in response to both cue types was also assessed parametrically. Brain activity during encoding which leads to different degrees of subsequent memory accuracy under the two retrieval conditions was compared. We found the bilateral posterior PHC to participate in encoding of both the object associated with a location and the location associated with an object. In contrast, activity in an area in the left anterior PHC and the right anterior MTL was only correlated with the memory for the location associated with an object.

OBJECTIVE: To evaluate the neuropsychological effects of current low level and previous higher levels of exposure to lead and evaluate the relation between effects of lead and bone lead. METHODS: A neuropsychological test battery was given to 54 storage battery workers with well documented long term exposure to lead. The effect was studied in two subgroups: those whose blood lead had never exceeded 2.4 mmol/l (the low BPbmax group, n = 26), and those with higher exposure about 10 years earlier (the high BPbmax group, n = 28). In both groups, the recent exposure had been low. Correlations between the test scores and the indices of both long term and recent exposure--including the content of lead in the tibial and calcaneal bone--and covariance analyses were used to assess the exposure-effect relation. Age, sex, and education were controlled in these analyses. RESULTS: Analyses within the low BPbmax group showed a decrement in visuospatial and visuomotor function (block design, memory for design, Santa Ana dexterity), attention (digit symbol, digit span), and verbal comprehension (similarities) associated with exposure to lead and also an increased reporting of subjective symptoms. The performance of the high BPbmax group was worse than that of the low BPbmax group for digit symbol, memory for design, and embedded figures, but there was no reporting of symptoms related to exposure, probably due to selection in this group. No relation was found between the output variables and the tibial lead concentration. The calcaneal lead concentrations were related to the symptoms in the low BPbmax group. CONCLUSIONS: Neuropsychological decrements found in subjects with high past and low present exposure indicate that blood lead concentrations rising to 2.5- 4.9 mmol/l cause a risk of long lasting or even permanent impairment of central nervous system function. Milder and narrower effects are associated with lower exposures; their reversibility and time course remain to be investigated. History of blood lead gives a more accurate prediction of the neuropsychological effects of lead than do measurements of bone lead.

Examined--by means of a Starkiewicz's hand localizer--were 170 right-hand workers of a paper factory working in conditions of excessive noise and high temperature of the surroundings. Analyzed were 140 persons tested 4 times in the period of each production shift (6.00-14.00; 14.00-22.00, and 22.00-6.00). The analysis showed the maximal efficacy of the visual function of the hand localization in the range of the right-hand and in the course of the night shift in both age groups.

The world of Machine Learning and Computer Vision has experienced a revolution since the last years. The appearance of Deep Learning algorithms and Convolutional Neural Networks, altogether with the increased processing capabilities provided by modern GPUs and the enormous amounts of annotated data publicly available, have allowed a boost in the field as never seen before. These notable improvements achieved in the Machine Learning world have led to the appearance of new fields like the Multimodal Learning, which encompasses and learns from many subfields. Additionally, new applications have taken profit of these advancements in order to reach high levels of performance. The huge results improvement of the currently available algorithms have allowed not only revolutionizing the academic world, but also bringing AI-based solutions to the market that looked like science fiction barely 10 years ago. This thesis, which is written as a papers compendium, focuses on delving deeper into the novel topic of Deep Multimodal Learning by proposing new algorithms and solutions for both already existing and newly defined problems. From the applications perspective, most of the papers presented can be divided in two areas of applicability. From the one hand, Egocentric Vision and Storytelling, which consists in acquiring images from the daily life of a person in order to analyse its behaviour patterns like social interactions, activities and events, interactions with objects, etc. And on the other hand, Food Recognition and Analysis, which consists in visually analysing and recognizing the food appearing on images in multiple contexts and with different levels of complexity, from food groups recognition to nutritional analysis. In both applications, the final purpose of the proposed papers is building tools that provide information that could lead to a better quality of life of the users. El mundo del Machine Learning y la Visión por Computador ha experimentado una revolución los últimos años. La aparición de algoritmos de Deep Learning y Convolutional Neural Networks, junto con las mayores capacidades de procesamiento proporcionadas por GPU modernas y las enormes cantidades de datos anotados disponibles públicamente, han permitió un impulso en el campo como nunca antes se había visto.Estas notables mejoras logradas en el mundo del Machine Learning han llevado a la aparición de nuevos campos como el Aprendizaje Multimodal, que engloba y aprende de muchos subcampos. Además, nuevas aplicaciones han aprovechado estos avances para alcanzar altos niveles de rendimiento. La enorme mejora en los resultados de los algoritmos disponibles actualmente ha permitido no solo revolucionar el mundo académico, sino también llevar al mercado soluciones basadas en IA que parecían ciencia ficción hace apenas 10 años.Esta tesis, que está escrita como un compendio de artículos, se enfoca en profundizar en el novedoso tema del Aprendizaje Multimodal Profundo al proponer nuevos algoritmos y soluciones para problemas ya existentes y recientemente definidos. Desde la perspectiva de las aplicaciones, la mayoría de los trabajos presentados se pueden dividir en dos áreas de aplicabilidad. Por un lado, la Visión Egocéntrica y el Storytelling, que consiste en la adquisición de imágenes de la vida diaria de una persona para analizar su comportamiento y extraer patrones asociadas a estos como por ejemplo interacciones sociales, actividades y eventos, interacciones con objetos, etc. Y por otro lado, el Reconocimiento y Análisis de Alimentos, que consiste en visualmente analizar y reconocer la comida que aparece en imágenes en múltiples contextos y con diferentes niveles de complejidad, desde el reconocimiento de grupos de alimentos hasta el análisis nutricional.En ambas aplicaciones, el propósito final de los artículos propuestos es construir herramientas que brinden información que pueda conducir a una mejor calidad de vida de los usuarios.

We have no difficulty seeing a straight line on a curved piece of paper, but in fact, doing so requires decoupling the shape of the surface from the pattern itself. Here we report a novel form of invariance in the visual cortex: single neurons in monkey inferior temporal cortex respond similarly to congruent transformations of patterns and surfaces, in effect decoupling patterns from the surface on which they are overlaid., We have no difficulty seeing a straight line drawn on a paper even when the paper is bent, but this inference is in fact nontrivial. Doing so requires either matching local features or representing the pattern after factoring out the surface shape. Here we show that single neurons in the monkey inferior temporal (IT) cortex show invariant responses to patterns across rigid and nonrigid changes of surfaces. We recorded neuronal responses to stimuli in which the pattern and the surrounding surface were varied independently. In a subset of neurons, we found pattern-surface interactions that produced similar responses to stimuli across congruent pattern and surface transformations. These interactions produced systematic shifts in curvature tuning of patterns when overlaid on convex and flat surfaces. Our results show that surfaces are factored out of patterns by single neurons, thereby enabling complex perceptual inferences. NEW & NOTEWORTHY We have no difficulty seeing a straight line on a curved piece of paper, but in fact, doing so requires decoupling the shape of the surface from the pattern itself. Here we report a novel form of invariance in the visual cortex: single neurons in monkey inferior temporal cortex respond similarly to congruent transformations of patterns and surfaces, in effect decoupling patterns from the surface on which they are overlaid.

This paper aims at advancing the field of electro-sensing. It exhibits the physical mechanism underlying shape perception for weakly electric fish. These fish orient themselves at night in complete darkness by employing their active electrolocation system. They generate a stable, high-frequency, weak electric field and perceive the transdermal potential modulations caused by a nearby target with different admittivity than the surrounding water. In this paper, we explain how weakly electric fish might identify and classify a target, knowing by advance that the latter belongs to a certain collection of shapes. Our model of the weakly electric fish relies on differential imaging, i.e., by forming an image from the perturbations of the field due to targets, and physics-based classification. The electric fish would first locate the target using a specific location search algorithm. Then it could extract, from the perturbations of the electric field, generalized (or high-order) polarization tensors of the target. Computing, from the extracted features, invariants under rigid motions and scaling yields shape descriptors. The weakly electric fish might classify a target by comparing its invariants with those of a set of learned shapes. On the other hand, when measurements are taken at multiple frequencies, the fish might exploit the shifts and use the spectral content of the generalized polarization tensors to dramatically improve the stability with respect to measurement noise of the classification procedure in electro-sensing. Surprisingly, it turns out that the first-order polarization tensor at multiple frequencies could be enough for the purpose of classification. A procedure to eliminate the background field in the case where the permittivity of the surrounding medium can be neglected, and hence improve further the stability of the classification process, is also discussed., Comment: 10 pages, 15 figures

The ability of 26 younger (mean age was 22.5 years) and older adults (mean age was 72.6 years) to haptically perceive material properties was evaluated. The participants manually explored (for 5 seconds) 42 surfaces twice and placed each of these 84 experimental stimuli into one of seven categories: paper, plastic, metal, wood, stone, fabric, and fur/leather. In general, the participants were best able to identify fur/leather and wood materials; in contrast, recognition performance was worst for stone and paper. Despite similar overall patterns of performance for younger and older participants, the younger adults’ recognition accuracies were 26.5% higher. The participants’ tactile acuities (assessed by tactile grating orientation discrimination) affected their ability to identify surface material. In particular, the Pearson r correlation coefficient relating the participants’ grating orientation thresholds and their material identification performance was −0.8: The higher the participants’ thresholds, the lower the material recognition ability. While older adults are able to effectively perceive the solid shape of environmental objects using the sense of touch, their ability to perceive surface materials is significantly compromised.

This paper proposes a biologically plausible matching method to recognize general shapes based on contour curvature information. The human visual system recognizes general shapes flexibly in real-world scenes through the ventral pathway. The pathway is typically modeled using artificial neural networks. These network models, however, do not construct a shape representation that satisfies the following required constraints: (1) The original shape should be represented by a group of partitioned contours in order to retrieve the whole shape (global information) from the partial contours (local information). (2) Coarse and fine structures of the original shapes should be individually represented in order for the visual system to respond to shapes as quickly as possible based on the least number of their features, and to discriminate between shapes based on detailed information. (3) The shape recognition realized with an artificial visual system should be invariant to geometric transformation such as expansion, rotation, or shear. In this paper, we propose a visual shape representation with geometrically characterized contour partitions described on multiple spatial scales.

Although asymmetric colour matching has been widely used in experiments on colour constancy, an exact colour match between objects lit by different chromatic lights is impossible to achieve. We used a modification of this technique, instructing our observers to establish the least dissimilar pair of differently illuminated coloured papers. The stimulus display consisted of two identical sets of 22 Munsell papers illuminated independently by neutral, yellow, blue, green and red lights. The lights produced approximately the same illuminance. Four trichromatic observers participated in the experiment. The proportion of exact matches was evaluated. When both sets of papers were lit by the same light, the exact match rate was 0.92, 0.93, 0.84, 0.78 and 0.76 for the neutral, yellow, blue, green and red lights, respectively. When one illumination was neutral and the other chromatic, the exact match rate was 0.80, 0.40, 0.56 and 0.32 for the yellow, blue, green and red lights, respectively. When both lights were chromatic, the exact match rate was found to be even poorer (0.30 on average). Yet, least dissimilar matching was found to be rather systematic. Particularly, a statistical test showed it was symmetric and transitive. The exact match rate was found to be different for different papers, varying from 0.99 (black paper) to 0.12 (purple paper). Such a variation can hardly be expected if observers' judgements were based on an illuminant estimate. We argue that colour constancy cannot be achieved for all the reflecting objects because of mismatching of metamers. We conjecture that the visual system might have evolved to have colour constant perception for some ecologically valid objects at a cost of colour inconstancy for other types of objects.

Object recognition is one of the most important functions of the human visual system, yet one of the least understood, this despite the fact that vision is certainly the most studied function of the brain. We understand relatively well how several processes in the cortical visual areas that support recognition capabilities take place, such as orientation discrimination and color constancy. This paper proposes a model of the development of object recognition capability, based on two main theoretical principles. The first is that recognition does not imply any sort of geometrical reconstruction, it is instead fully driven by the two dimensional view captured by the retina. The second assumption is that all the processing functions involved in recognition are not genetically determined or hardwired in neural circuits, but are the result of interactions between epigenetic influences and basic neural plasticity mechanisms. The model is organized in modules roughly related to the main visual biological areas, and is implemented mainly using the LISSOM architecture, a recent neural self-organizing map model that simulates the effects of intercortical lateral connections. This paper shows how recognition capabilities, similar to those found in brain ventral visual areas, can develop spontaneously by exposure to natural images in an artificial cortical model.

The paper describes a new adaptive neural network for edge-based pattern extraction problems. In the model, each neuron represents an edge with continuous state variables describing its location and orientation. The post-synaptic distribution is dependent on the state variables. Each neuron adjusts its state to increase its membrane potential, which results in highly adaptive dynamics of the synaptic weight distribution. The network allocates multiple neurons with different orientation modes for each edge. The strategy allows accurate modeling of multi-modal distributions at key-points such as corners and junctions. As a result, the network delineates edges at sub-pixel accuracy while preserving key-points. It is also capable of processing a sequence of images and following moving objects. The network is extended for contour integration and key-point detection tasks. The paper presents experiments conducted on both synthetic and non-synthetic data to demonstrate the effectiveness of the technique.

This paper uses visual, empirical and formal methods (Li & Zaidi, Vision Research, 40 (2000) 217; Li & Zaidi, Vision Research, 41 (22) (2001a) 2927) to examine the roles of oriented texture components in conveying veridical percepts of concave and convex surfaces that are pitched towards or away from the observer. The results show that pairs of components, oriented symmetrically around the axis of maximum curvature, combine to provide the geodesic orientation modulations that are critical for veridical shape perception. The degree of pitch determines the orientations of the critical pair of components. Perspective is crucial to the veridical perception of concavities and convexities, regardless of the degree of pitch. The results of this paper reconfirm that veridical shape perception depends on extracting critical patterns of oriented energy, but also show that the class of textures capable of conveying veridical percepts of developable shapes in general views is even more restricted than that identified by Li and Zaidi (Journal of Optical Society of America A, 18 (2001b), 2430).

The modern study of perception began when Fechner published his ‘Elements of Psychophysics’ in 1860. This book has guided most perception research ever since. It has become increasingly clear that there are problems with Fechner's approach, which assumes that the percept is completely determined by the sensory input. Fechner's approach cannot explain the processes that allow our percepts to be veridical. Post-Fechnerian schools (Helmholtzian, Structural, Gestalt and Gibsonian) have tried to deal with this problem, but have not been successful. An alternative to the Fechnerian approach is required. This paper describes an alternative that has been developing over the last 20 years within the computer vision community. It treats perceptual interpretation as a solution of an inverse problem that depends critically on the operation of a priori constraints. Contemporary research, which adopted this approach, has concentrated on verifying the usefulness of Bayesian and standard regularization methods. This paper takes the next step; it discusses theoretical and empirical aspects of studying human perception as an inverse problem. It reviews the literature that illustrates the power of the inverse problem approach. This review leads to the suggestion that progress in the study of perception will benefit if the inverse approach were to be adopted by experimentalists, as well as by the computational modelers, who have been actively exploring its potential to date.

This and the following two papers describe event-related potentials (ERPs) evoked by visual stimuli in 98 patients in whom electrodes were placed directly upon the cortical surface to monitor medically intractable seizures. Patients viewed pictures of faces, scrambled faces, letter-strings, number-strings, and animate and inanimate objects. This paper describes ERPs generated in striate and peristriate cortex, evoked by faces, and evoked by sinusoidal gratings, objects and letter-strings. Short-latency ERPs generated in striate and peristriate cortex were sensitive to elementary stimulus features such as luminance. Three types of face-specific ERPs were found: (i) a surface-negative potential with a peak latency of approximately 200 ms (N200) recorded from ventral occipitotemporal cortex, (ii) a lateral surface N200 recorded primarily from the middle temporal gyrus, and (iii) a late positive potential (P350) recorded from posterior ventral occipitotemporal, posterior lateral temporal and anterior ventral temporal cortex. Face-specific N200s were preceded by P150 and followed by P290 and N700 ERPs. N200 reflects initial face-specific processing, while P290, N700 and P350 reflect later face processing at or near N200 sites and in anterior ventral temporal cortex. Face-specific N200 amplitude was not significantly different in males and females, in the normal and abnormal hemisphere, or in the right and left hemisphere. However, cortical patches generating ventral face-specific N200s were larger in the right hemisphere. Other cortical patches in the same region of extrastriate cortex generated grating-sensitive N180s and object-specific or letter-string-specific N200s, suggesting that the human ventral object recognition system is segregated into functionally discrete regions.

This is the last in a series of papers dealing with intracranial event-related potential (ERP) correlates of face perception. Here we describe the results of manipulations that may exert top-down influences on face recognition and face-specific ERPs, and the effects of cortical stimulation at face-specific sites. Ventral face-specific N200 was not evoked by affective stimuli; showed little or no habituation; was not affected by the familiarity or unfamiliarity of faces; showed no semantic priming; and was not affected by face-name learning or identification. P290 and N700 were affected by semantic priming and by face-name learning and identification. The early fraction of N700 and face-specific P350 exhibited significant habituation. About half of the AP350 sites exhibited semantic priming, whereas the VP350 and LP350 sites did not. Cortical stimulation evoked a transient inability to name familiar faces or evoked face-related hallucinations at two-thirds of face-specific N200 sites. These results are discussed in relation to human behavioral studies and monkey single-cell recordings. Discussion of results of all three papers concludes that: face-specific N200 reflects the operation of a module specialized for the perception of human faces; ventral and lateral occipitotemporal cortex are composed of a complex mosaic of functionally discrete patches of cortex of variable number, size and location; in ventral cortex there is a posterior-to-anterior trend in the location of patches in the order letter-strings, form, hands, objects, faces and face parts; P290 and N700 at face-specific N200 sites, and face-specific P350, are subject to top-down influences.

In the previous paper the locations and basic response properties of N200 and other face-specific event-related potentials (ERPs) were described. In this paper responsiveness of N200 and related ERPs to the perceptual features of faces and other images was assessed. N200 amplitude did not vary substantially, whether evoked by colored or grayscale faces; normal, blurred or line-drawing faces; or by faces of different sizes. Human hands evoked small N200s at face-specific sites, but evoked hand-specific ERPs at other sites. Cat and dog faces evoked N200s that were 73% as large as to human faces. Hemifield stimulation demonstrated that the right hemisphere is better at processing information about upright faces and transferring it to the left hemisphere, whereas the left hemisphere is better at processing information about inverted faces and transferring it to the right hemisphere. N200 amplitude was largest to full faces and decreased progressively to eyes, face contours, lips and noses viewed in isolation. A region just lateral to face-specific N200 sites was more responsive to internal face parts than to faces, and some sites in ventral occipitotemporal cortex were face-part-specific. Faces with eyes averted or closed evoked larger N200s than those evoked by faces with eyes forward. N200 amplitude and latency were affected by the joint effects of eye and head position in the right but not in the left hemisphere. Full and three-quarter views of faces evoked larger N200s than did profile views. The results are discussed in relation to behavioral studies in humans and single-cell recordings in monkeys.

The ganglion cells of the primate retina include two major anatomical and functional classes: P cells which project to the four parvocellular layers of the lateral geniculate nucleus (LGN), and M cells which project to the two magnocellular layers. The characteristics of the P-cell receptive field are central to understanding early form and color vision processing (Kaplan et al., 1990; Schiller & Logothetis, 1990). In this and in the following paper, P-cell dynamics are systematically analyzed in terms of linear and nonlinear response properties. Stimuli that favor either the center or the surround of the receptive field were produced on a CRT and modulated with a broadband signal composed of multiple m-sequences (Benardete et al., 1992b; Benardete & Victor, 1994). The first-order responses were calculated and analyzed in this paper (part I). The findings are: (1) The first-order responses of the center and surround depend linearly on contrast. (2) The dynamics of the center and surround are well described by a bandpass filter model. The most significant difference between center and surround dynamics is a delay of approximately 8 ms in the surround response. (3) In the LGN, these responses are attenuated and delayed by an additional 1–5 ms. (4) The spatial transfer function of the P cell in response to drifting sine gratings at three temporal frequencies was measured. This independent method confirmed the delay between the (first-order) responses of the center and surround. This delay accounts for the dependence of the spatial transfer function on the frequency of stimulation.

Adult subjects drew the visual projection of two models. One model was a trapezoid placed in the frontoparallel plane. The other was a tilted rectangle which displayed the same projective shape on a frontoparallel plane as the trapezoid. The drawing conditions were varied in two ways: the model remained available for inspection during the drawing task or it was masked after initial inspection; the subjects drew on paper placed flat on the table or on a vertical glass pane placed in front of the model (ie on a da Vinci window). The results were that (i) the projective shape of the frontoparallel trapezoid was reproduced accurately whereas that of the tilted rectangle was systematically distorted in the direction of its actual physical dimensions; (ii) when subjects drew on paper, the presence or absence of a view of the model made no difference to the amount of distortion; (iii) drawing on a da Vinci window improved accuracy even when the model was hidden. These findings provide information about the relative roles of object-centred knowledge, perceptual abilities, and depiction skills in drawing performance.

A patient with severe left hemi-inattention was found to be influenced by neglected stimuli in free drawing tasks. He was presented 64 sheets of paper: 32 had geometric figures on the left side and 32 on the right side. The patient was asked what he saw and invited to draw a subject of his choice on the sheet of paper. The patient always denied seeing anything when the figures were on the left side, while he identified the figures on the right side. Four judges examined the patient's drawings and all of them agreed that 11 drawings were influenced by neglected stimuli and none by non-neglected ones. The drawings he made were similar but not the exact reproduction of neglected figures. This finding confirms that visual processing can proceed without awareness in a patient with hemi-inattention.

Previous studies of the neural mechanisms of roughness perception have provided evidence that the magnitude of perceived roughness depends on spatial variation in the afferent population discharge. This paper reviews those studies and provides new data that appear to show that roughness perception cannot depend on activity in cutaneous rapidly adapting afferent fibers when surface element spacings exceed 1 mm. Finer surfaces have not been studied in combined psychophysical and neurophysiological studies. This paper also reinterprets the data of an earlier study, showing that all within-fiber neural coding mechanisms, which include impulse rate codes and codes based on the temporal structure of the afferent signal, can be rejected as the basis for roughness perception when the finger scans a textured surface.Key words: roughness, texture, somatosensory, psychophysics, neurophysiology.

In this paper a self organising map is proposed for object recognition based on tactile form perception. A robot hand with three fingers, with the same number of degrees of freedom as the human hand, is used for obtaining the required tactile measurements. Finger joint angles were recorded when the hand was grasping different objects, in three different orientations. A self organising map was used to categorise objects based on the measurements obtained using the hand. The proposed system learnt to recognise objects of different shape, as well as objects of the same shape but different size. To test the generalisation ability of the system, new objects (different from the training set) were applied and it was observed that the system learnt to categorise objects based on their shape and size. This paper also investigates the reliability of object recognition using tactile form perception with noisy measurements, the maximum number of different objects that can be recognised without significantly degrading the performance and the ability to recognise similar shaped objects with small differences in dimensions. Based on the test results presented, the system can recognise 89 % of 25 different objects. This promising performance suggests that tactile form perception can be reliably used for object recognition in robotic applications.

In this paper an attempt is made to review the types of processes developed by the author, and others, to classify and segment textures and to compare, where possible, such algorithms with what has been proposed to occur in human vision. In particular, this paper is concerned with evaluating the proposed constituent processes from a 'cognitive engineering' perspective where, in order to have an adequate model for either human or machine texture processing, complete segmentation or classification or both must be attained.

Observers can easily differentiate between a pigmented stain and the white surface that it lies on. The same applies for a colour shadow cast upon the same surface. Although the difference between these two kinds of colour appearance (referred to as material and lighting hues) is self-evident even for inexperienced observers, it is not one that has been captured by any colour appearance model thus far. We report here on an experiment supplying evidence for the dissociation of these two types of hue in the perceptual space. The stimulus display consisted of two identical sets of Munsell papers illuminated independently by yellow, neutral, and blue lights. Dissimilarities between all the paper/light pairs were ranked by five trichromatic observers, and then analysed by using non-metric multidimensional scaling (MDS). In the MDS output configuration, the Munsell papers lit by the same light made a closed configuration retaining the same order as in the Munsell book. The paper configurations for the yellow and blue lights were displaced transversally and in parallel to each other, with that of the neutral light located in between. The direction of the shift is interpreted as the yellow-blue lighting dimension. We show that the yellow-blue lighting dimension cannot be reduced to that of the reflected light.

We used a novel probe disk technique to test for the existence of functional frames of reference for lightness perception in complex images. Thirteen identical gray disks were electronically pasted into the photograph Trastevere, which shows two large regions of sunlight and shadow. Observers matched the lightness of each disk with a Munsell scale. The data revealed a framework effect. That is, lightness differences within either the sunlight or shadow region were small relative to the pronounced step function at the framework boundary. Additional experiments testing the perceived embeddedness of the disks showed that the framework effect was increased when disk size and shape were altered to conform to the perspective shown in the photograph and when the disks were blurred slightly to conform to the graininess of the photograph. The effect was further increased when the photograph was viewed through a pinhole and when the disks were presented one by one. The effect was reduced when paper disks of equal luminance and visual angle were pasted onto the glass front of the CRT screen. When the sunlight framework was covered with black paper, the remained disks within the shadow region appeared white, as predicted by the anchoring theory (A. Gilchrist, 2006).

Harman et al (1999, Current Biology 9 1315 – 1318) showed that active exploration of novel objects allows faster recognition of these objects than passive viewing. Here, we examined whether active exploration of particular objects can facilitate subsequent view matching of different objects within the same category. In the generalisation phase, participants performed a matching task between two serially presented views of novel paper-clip objects. When the two views were of the same object, these were related by rotation about the vertical or horizontal axis. Subsequently, in the observation phase, participants were presented with another 5 paper-clip objects. One group of participants (active group) actively explored each of the objects for 20 s using a trackball over a limited range (-45° to +45°) of the horizontal axis. The other participants observed a replay of the active exploration of each of these objects by a participant in the active group (passive group) or observed a static view of the objects (still group). Following this observation phase, the generalisation phase was repeated twice for all groups of participants. The active group showed a clear expansion of the generalisation range about the horizontal axis. However, this expansion disappeared when we replicated the experiment using a smaller range of active exploration (-30° to +30°) and the experiment in which the participants observed a rotating object within the range of -45° to +45°. Furthermore, we found a significant improvement in view generalisation about the vertical axis when participants actively explored objects using a turntable over a range in -45° to +45° of the vertical axis. Most importantly, since participants explored different objects from the ones that were used in the generalisation phase, these results cannot be explained by improvement of the encoding process. The present findings therefore suggest that: (i) active exploration facilitates the process of view matching, (ii) this effect depends on the exploration range and rotational axis, and (iii) interaction between motor and visual information is necessary to facilitate the process of view matching., Sasaoka T., Asakura N., Kawahara T., 2010. The definitive, peer-reviewed and edited version of this article is published in Perception 39(3), 289-308 2010, 10.1068/p5721

Twenty five patients with right cerebral hemisphere damage and neglect participated in a series of bisection experiments. As expected, long lines were bisected to the right of true midpoint. By contrast, large circles and long white paper strips were bisected accurately, or with leftward errors. Small objects were less sensitive to stimulus properties: short lines and paper strips, and small circles, were bisected to the left of true midpoint, and these leftward errors were equally common as rightward errors with long lines. When asked to draw a perpendicular line of the same length as the presented horizontal line, patients overestimated the length of short lines but underestimated that of long lines. Presenting lines in near and far extrapersonal space selectively affected bisection of short lines. The results suggest that two opposing, independent mechanisms determine bisection performance in left neglect.

A computerized version of the Revised Mental Rotations Test using touch-screen technology was tested and compared with the paper-and-pencil version. One hundred ninety-two participants--133 women and 59 men--took the paper-and-pencil version; 91 participants--47 women and 44 men--took the touch-screen version. Standard scores x test half and the entire test, proportion of errors x type, and proportion correct x alternative type were calculated. The expected gender differences occurred in standard scores and proportion correct x alternative type for both test versions. Men performed better than women in all instances, but gender difference effect sizes (measured by Cohen's d) were reduced from large for the paper-and-pencil version to medium for the computerized version. Scores declined at least nominally from the first to the second half of both versions, and significantly for women taking the paper-and-pencil version.

Visual context between objects is an important cue for object position perception. How to effectively represent the visual context is a key issue to study. Some past work introduced task-driven methods for object perception, which led a large coding quantity. This paper proposes an approach that incorporates feature-driven mechanism into object-driven context representation for object locating. As an example, the paper discusses how a neuronal network encodes the visual context between feature salient regions and human eye centers with as little coding quantity as possible. A group of experiments on efficiency of visual context coding and object searching are analyzed and discussed, which show that the proposed method decreases the coding quantity and improve the object searching accuracy effectively.

Humans typically have a remarkable memory for faces. Nonetheless, in some cases they can be fooled. Experiments described in this paper provide new evidence for an effect in which observers falsely “recognize” a face that they have never seen before. The face is a chimera (prototype) built from parts extracted from previously viewed faces. It is known that faces of this kind can be confused with truly familiar faces, a result referred to as the prototype effect. However, recent studies have failed to find evidence for a full effect, one in which the prototype is regarded not only as familiar, but as more familiar than faces which have been seen before. This study sought to reinvestigate the effect. In a pair of experiments, evidence is reported for the full effect based on both an old/new discrimination task and a familiarity ranking task. The results are shown to be consistent with a recognition model in which faces are represented as combinations of reusable, abstract features. In a final experiment, novel predictions of the model are verified by comparing the size of the prototype effect for upright and upside-down faces. Despite the fundamentally piecewise nature of the model, an explanation is provided as to how it can also account for the sensitivity of observers to configural and holistic cues. This discussion is backed up with the use of an unsupervised network model. Overall, the paper describes how an abstract feature-based model can reconcile a range of results in the face recognition literature and, in turn, lessen currently perceived differences between the representation of faces and other objects.

The study of visual perception in human infants is confronted by a number of special problems arising from the inaccessibility of verbal reports. In this paper, we discuss the experimental strategy of converging operations in the context of investigating the phenomenal experience of infants. The goal of this strategy is to logically and empirically delimit alternative explanations for a given behavior. Knowledge about the mature functioning of a perceptual competence, as well as knowledge about its developmental course, constrains the selection of viable explana­ tions, but cannot produce a unique interpretation. This goal is pursued through the implementa­ tion of an iterative strategy in which competing interpretations are tested until only one plausi­ ble alternative remains. A series of experiments investigating infants' sensitivity to biomechanical motions are reviewed as a way of illustrating how this methodology is operationalized. The perceptual world is formed by proximal stimula­ tion and perceptual processes that are (I) shaped over the evolutionary course of our species or (2) acquired dur­ ing the individual's developmental history. By definition, the human infant's perceptual world is primarily struc­ tured by innate perceptual processes, whereas the percep­ tual world of the adult benefits from extensive experience and knowledge about the environment. It thus follows that one of the most effective strategies for assessing basic competencies in perception is to study young infants. What can we know of babies' perceptual experience? Skeptics argue that we can never know about the phe­ nomenal experiences of infants because our assessments are restricted to nonverbal responses. By this reasoning, perceptual research with animals, as well as with human infants, yields results that will always fall short of those obtained with older children and adults who can provide verbal phenomenal reports. Contrary to this argument, the principle premise of this paper is that empirical evi­ dence pertaining to babies' phenomenal experience is, in­ deed, attainable, even though the inaccessibility of ver­ bal reports presents special problems in infant perception research.

Recent solutions to object classification have focused on the decomposition of objects into representative parts. However, the vast majority of these methods are based on single visual cue measurements. Psychophysical evidence suggests that humans use multiple visual cues to accomplish recognition. In this paper, we address the problem of integrating multiple visual information for object recognition. Our contribution in this paper is twofold. First, we describe a new probabilistic integration model of multiple visual cues at different spatial locations across the image. Secondly, we use the cue integration framework to classify images of objects by combining two-dimensional and three-dimensional visual cues. Classification results obtained using the method are promising.

Shape complexity has recently received attention from different fields, such as computer vision and psychology. In this paper, integral geometry and information theory tools are applied to quantify the shape complexity from two different perspectives: from the inside of the object, we evaluate its degree of structure or correlation between its surfaces (inner complexity), and from the outside, we compute its degree of interaction with the circumscribing sphere (outer complexity). Our shape complexity measures are based on the following two facts: uniformly distributed global lines crossing an object define a continuous information channel and the continuous mutual information of this channel is independent of the object discretisation and invariant to translations, rotations, and changes of scale. The measures introduced in this paper can be potentially used as shape descriptors for object recognition, image retrieval, object localisation, tumour analysis, and protein docking, among others.

The Hermann grid illusion consists of smudges perceived at the intersections of a white grid presented on a black background. In 1960 the effect was first explained by a theory advanced by Baumgartner suggesting the illusory effect is due to differences in the discharge characteristics of retinal ganglion cells when their receptive fields fall along the intersections versus when they fall along non-intersecting regions of the grid. Since then, others have claimed that this theory might not be adequate, suggesting that a model based on cortical mechanisms is necessary [Lingelbach et al, 1985 Perception14(1) A7; Spillmann, 1994 Perception23 691–708; Geier et al, 2004 Perception33 Supplement, 53; Westheimer, 2004 Vision Research44 2457–2465]. We present in this paper the following evidence to show that the retinal ganglion cell theory is untenable: (i) varying the makeup of the grid in a manner that does not materially affect the putative differential responses of the ganglion cells can reduce or eliminate the illusory effect; (ii) varying the grid such as to affect the putative differential responses of the ganglion cells does not eliminate the illusory effect; and (iii) the actual spatial layout of the retinal ganglion cell receptive fields is other than that assumed by the theory. To account for the Hermann grid illusion we propose an alternative theory according to which the illusory effect is brought about by the manner in which S1 type simple cells (as defined by Schiller et al, 1976 Journal of Neurophysiology39 1320–1333) in primary visual cortex respond to the grid. This theory adequately handles many of the facts delineated in this paper.

This study examined how children used shape similarity information when they inferred the meaning of a novel label which was given to a familiar object. When children were shown two objects similar only in shape (white paper cup and glass), 5 : 7- to 6 : 7-year-old children (N = 28, M = 5 : 11) assigned the novel label only to the object identical to the target, whereas 4 : 7- to 5 : 6-year-old children (N = 28, M 4 : 11) mapped the novel label to all objects within the familiar object category. When children of both age groups (N = 15, M = 5 : 8, range = 5: 3 to 6 : 2; N 12, M = 4 : 10, range = 4 : 3 to 5 : 1) were shown two paper cups identical except color (white paper cup and pink one), children selected both of the two objects. These results showed children in both age groups regarded the objects identical except color as those given the same label; on the other hand, children did not think of the objects similar only in shape as qualified for the same label. Therefore the results imply that what is treated as "shape similarity" in this area of research includes the similarity not only in shape but also in other dimensions such as material.

In recent years the problem of object recognition has received considerable attention from both the machine learning and computer vision communities. The key challenge of this problem is to be able to recognize any member of a category of objects in spite of wide variations in visual appearance due to variations in the form and colour of the object, occlusions, geometrical transformations (such as scaling and rotation), changes in illumination, and potentially non-rigid deformations of the object itself. In this paper we focus on the detection of objects within images by combining information from a large number of small regions, or ‘patches', of the image. Since detailed hand-segmentation and labelling of images is very labour intensive, we make use of ‘weakly labelled' data in which the training images are labelled only according to the presence or absence of each category of object. A major challenge presented by this problem is that the foreground object is accompanied by widely varying background clutter, and the system must learn to distinguish the foreground from the background without the aid of labelled data. In this paper we first show that patches which are highly relevant for the object discrimination problem can be selected automatically from a large dictionary of candidate patches during learning, and that this leads to improved classification compared to direct use of the full dictionary. We then explore alternative techniques which are able to provide labels for the individual patches, as well as for the image as a whole, so that each patch is identified as belonging to one of the object categories or to the background class. This provides a rough indication of the location of the object or objects within the image. Again these individual patch labels must be learned on the basis only of overall image class labels. We develop two such approaches, one discriminative and one generative, and compare their performance both in terms of patch labelling and image labelling. Our results show that good classification performance can be obtained on challenging data sets using only weak training labels, and they also highlight some of the relative merits of discriminative and generative approaches.