We investigated the cortical mechanisms underlying the visual perception of luminance-defined surfaces and the preference for black over white stimuli in the macaque primary visual cortex, V1. We measured V1 population responses with voltage-sensitive dye imaging in fixating monkeys that were presented with white or black squares of equal contrast around a mid-gray. Regions corresponding to the squares' edges exhibited higher activity than those corresponding to the center. Responses to black were higher than to white, surprisingly to a much greater extent in the representation of the square's center. Additionally, the square-evoked activation patterns exhibited spatial modulations along the edges and corners. A model comprised of neural mechanisms that compute local contrast, local luminance temporal modulations in the black and white directions, and cortical center-surround interactions, could explain the observed population activity patterns in detail. The model captured the weaker contribution of V1 neurons that respond to positive (white) and negative (black) luminance surfaces, and the stronger contribution of V1 neurons that respond to edge contrast. Also, the model demonstrated how the response preference for black could be explained in terms of stronger surface-related activation to negative luminance modulation. The spatial modulations along the edges were accounted for by surround suppression. Overall the results reveal the relative strength of edge contrast and surface signals in the V1 response to visual objects.

The neuronal mechanism underlying the representation of color surfaces in primary visual cortex (V1) is not well understood. We tested on color surfaces the previously proposed hypothesis that visual perception of uniform surfaces is mediated by an isomorphic, filled-in representation in V1. We used voltage-sensitive-dye imaging in fixating macaque monkeys to measure V1 population responses to spatially uniform chromatic (red, green, or blue) and achromatic (black or white) squares of different sizes (0.5°-8°) presented for 300 ms. Responses to both color and luminance squares early after stimulus onset were similarly edge-enhanced: for squares 1°and larger, regions corresponding to edges were activated much more than those corresponding to the center. At later times after stimulus onset, responses to achromatic squares' centers increased, partially "filling-in" the V1 representation of the center. The rising phase of the center response was slower for larger squares. Surprisingly, the responses to color squares behaved differently. For color squares of all sizes, responses remained edge-enhanced throughout the stimulus. There was no filling-in of the center. Our results imply that uniform filled-in representations of surfaces in V1 are not required for the perception of uniform surfaces and that chromatic and achromatic squares are represented differently in V1.Key words: color; monkeys; population coding; primary visual cortex; surfaces; VSDI Introduction "… space and color are not distinct elements but, rather, are interdependent aspects of a unitary process of perceptual organization." (Kanizsa, 1979).The above quotation from Kanizsa's (1979) book guides our work on the neural basis of color perception. The brain needs to construct a color signal to recover the reflective properties of surfaces. Therefore, the neural mechanisms of color perception must make comparisons of the color signals from different locations in the visual image; they must take into account the spatial layout of the scene (Delahunt and Brainard, 2004; Shevell and Kingdom, 2008). It is not known yet in detail how the brain integrates form and color but many scientists who investigated the problem concluded that the primary visual cortex (V1) plays an important role (Johnson et al., 2001(Johnson et al., , 2008 Friedman et al., 2003;Wachtler et al., 2003; Hurlbert and Wolf, 2004).Many investigators have reported the existence of colorresponsive neurons in V1 of macaque monkeys (Thorell et al., 1984; Victor et al., 1994; Leventhal et al., 1995;Johnson et al., 2001; Friedman et al., 2003). Most of the color-sensitive neurons in V1 are double-opponent cells; they are orientation-tuned and respond best to intermediate spatial frequency gratings or to sharp edges in the visual image (ϳ30 -40% of V1 cells). Doubleopponent cells were shown to be sensitive to achromatic luminance patterns as well as to color patterns. Single-opponent cells Received March 29, 2015; revised July 18, 2015; accepted July 22, 2015. Author contributions: S.Z., R.S....

Sparse-to-dense interpolation for optical flow is a fundamental phase in the pipeline of most of the leading optical flow estimation algorithms. The current state-of-theart method for interpolation, EpicFlow, is a local average method based on an edge aware geodesic distance. We propose a new data-driven sparse-to-dense interpolation algorithm based on a fully convolutional network. We draw inspiration from the filling-in process in the visual cortex and introduce lateral dependencies between neurons and multilayer supervision into our learning process. We also show the importance of the image contour to the learning process. Our method is robust and outperforms EpicFlow on competitive optical flow benchmarks with several underlying matching algorithms. This leads to state-of-the-art performance on the Sintel and KITTI 2012 benchmarks.Inspired by neuroscience, we also suggest a loss involving lateral dependencies. Here, too, we replace the process of lateral feedback during run-time with additional supervision during training. In this way, the feedforward network learns how to mimic a network with lateral feedback loops by utilizing the training labels.Taken together, our contributions are: (a) We propose, for the first time, to the best of our knowledge, a neural network based sparse-to-dense interpolation for optical flow. Our network performs better than the current state of the art, it is robust and can be adjusted to different matching algorithms and serve as the new default interpolation method in optical flow pipelines. (b) We introduce a new lateral dependency loss, embedding the correlations between neighbors into the learning process. (c) We define a novel architecture involving detour networks in each layer of the network. The new architecture provides a substantial increase 1 arXiv:1611.09803v3 [cs.CV]

No abstract