41st European Conference on Visual Perception ECVP 2018
41st European Conference on Visual Perception ECVP 2018, Trieste, Italy
The classic Saliency Model by Itti and Koch launched many studies that contributed to the modelling of layers for vision and visual attention. The aim of this study is to improve the existing saliency model by using a neural network to generate salience maps to model human saccade generation. The proposed model uses a Leaky Integrate-and-Fire layer for temporal predictions, and replaces spatial salience with a deep learning neural network in order to create a generative model that combines spatial and temporal predictions. The results involve a deep neural network which is able to predict eye movements based on unsupervised learning from raw image input, as well as supervised learning from fixation maps retrieved during an eye-tracking experiment with 35 participants at later stages in order to train a 2D softmax layer. The results imply that it is possible to match model human fixation locations but temporal distributions are still limited by the accuracy of the leaky algorithm.
The allocation of attention can occur not only in space, but also in time. Application of Rescorla's "truly random control" procedure about independency of cues and targets allowed us to differentiate the impact of endogenous (voluntary) and exogenous (automatic) components of temporal attention on the performance separately and within their interaction. In a random dot motion task, variation of luminance and motion of dots, that represent the cue, affects the engagement of exogenous mode. Temporal contingency between cues and targets or its absence affects the impact of endogenous mode. Combining these conditions, the results are as follows. For endogenous cues, we see improvement of both speed and accuracy at early cue target onset asynchrony. For exogenous cues, we see improvement of response times, but not accuracy. When both are involved, we observe a trade-off of speed and accuracy. This parallels from the auditory modalities of alertness cueing but with purely visual stimuli.
As a foraging facilitator, Inhibition of return (IOR) must be coded in spatiotopic coordinates. Early reports confirmed this suggestion but these results have been recently challenged. The present study was designed to examine the reference frame of IOR and to test whether retinotopic IOR might be a part of the spatiotopic IOR gradient. We conducted four experiments with spatiotopically and retinotopically cued coordinates and an intervening saccade between the cue and target presentations. We alternated the response modality (manual and saccadic) and the cue-target spatial distance (fixed and contiguous). Our data showed evidence for an independent source of retinotopic IOR neither at discrete locations nor as a gradient; moreover, we observed the spread of IOR across the whole validly cued hemifield. We propose that these results indicate a strategy to attend and then inhibit the entire cued hemifield.
One of the important sources of failures in visual working memory (VWM) is that individual items can interfere with each other. Here, we tested how two causes of such interference—poor categorical distinctiveness and imperfect feature binding—interact. In three experiments, we showed low and high distinctive objects and tested VWM for objects alone, for locations alone and for object-location conjunctions. We found that low object distinctiveness impairs object recognition and increases the number of object-location binding errors. Also, we dissociated the probabilities that these binding errors are due to recognition impairment or a failure of correct binding. Results show that poor distinctiveness increases binding errors rate only due to lacking recognition but not to binding impairment. Together, our findings suggest that object distinction and object-location binding act upon different components of VWM and are separate sources of interference. This study was funded by RSCF #18-18-00334.
The adaptation aftereffect (AAE) of mean size suggests that mean size is coded as a basic visual property. Also, size-distance rescaling of individual objects occurs prior to averaging. Because it is unclear whether the AAE is based on rescaled mean size, we tested the degree of AAE as a function the apparent mean size of stimuli presented at different depths. Observers were stereoscopically shown an adapting patch of dots with either a large or small mean size, followed by a brief test circle. Adaptors and tests were presented at a near and a far plane, both in the same or in different planes. Observers then adjusted the size of a probe in the middle plane to match the test size. We found evidence of the AAE and for test size rescaling, but no effect of whether the adaptor and test were presented in the same or in different planes. Our results suggest that the AAE of mean size take places at a lower level of visual processing than size-distance rescaling. This study was funded by RFBR #18-313-00253.