Virtual Retina is a simulation software developped in the Odyssée research team (INRIA Sophia Antipolis - Méditerannée) by Adrien Wohrer (PhD, supervised by Pierre Kornprobst and Thierry Viéville). Virtual Retina allows large-scale simulations of biologically-plausible retinas, with customizable parameters, and different possible biological features: 1. Spatio-temporal linear filter implementing the basic Center/Surround organization of retinal filtering. 2. Non-linear contrast gain control mechanism providing instantaneous adaptation to the local level of contrast. This stage is modelled through dynamical adaptation conductances in the membranes of bipolar cells; the resulting model reproduces contrast-dependent amplitude and phase non-linearities, as measured in real mammalian retinas by Shapley & Victor 78. 3. Spike generation by one or several layers of ganglion cells paving the visual field. Magnocellular and Parvocellular pathways can be modelled in the same framework according to the parameters chosen. Large-scale simulations can be pursued on up to 100,000 spiking cells. 4. Possibility of a global radial inhomogeneity modeling the foveated organization of mammalian retinas. In this case, the spatial scales of filtering, and the density of spiking cells, both depend on the eccentricity from the center of the retina. 5. Possibility to include a basic microsaccades generator at the input of the retina, to account for fixational eye movements. Virtual Retina is under INRIA CeCill C open-source licence (IDDN number IDDN.FR.001.210034.000.S.P.2007.000.31235), so that you can download it, install it and run it on your own sequences. Virtual Retina also offers you a web service, so that you may test directly the main software on your own data, without any installation. This webservice was developed in collaboration with Nicolas Debeissat. Virtual Retina is the result of a research contribution, and you will also find the associated research publications if you want to learn more.

We develop a model of motion integration modulated by form information, inspired by neurobiological data. Our dynamical system models several key features of the motion processing stream in primate visual cortex. Thanks to a multi-layer architecture incorporating both feedforward and feedback and inhibitive lateral connections, our model is able to solve local motion ambiguities. One important feature of our model is to propose an anisotropic integration of motion based on the form information. Our model can be implemented efficiently on GPU and we show its properties on classical psychophysical examples. First, a simple read-out allows us to reproduce the dynamics of ocular following for a moving bar stimulus. Second, we show how our models able to discriminate between extrinsic and intrinsic junctions present in the chopstick illusion. We also obtain some promising results on real videos.

We propose a bio-inspired feedforward model for motion processing based on the neurophysiology literature, and we show how the estimated motion representation can be successfully used to recognize actions in real videos. As a major difference with the model proposed by Jhuang et al. (2007), we do really focus on proposing a model of motion processing which reproduces some key elements of the visual system in terms of dynamics and connectivity between the V1 and MT layers. In particular, we model the richness of center surround interactions in MT, arising from the integration of motion from the V1 cells. As it is observed in neurophysiology, the cells in our MT model not only behave like simple velocity detectors, but also respond to several kinds of motion contrasts. Interestingly, we show that this diversity of motion representation at the MT level is a major advantage for an action recognition task. We compare our results for action recognition on the Weizmann database, and show the performance of our approach with respect to approaches based on simply classical motion detectors.

We propose a simple mechanism to reproduce the neural solution of the aperture problem in monkey area MT. More precisely, our goal is to propose a model able to reproduce the dynamical change of the preferred direction (PD) of a MT cell depending on the motion information contained in the input stimulus. The PD of a MT cell measured through drifting gratings differs of the one measured using a barberpole, which is highly related with its aspect ratio. For a barberpole, the PD evolves from the perpendicular direction of the drifting grating to a PD shifted according to the aspect ratio of the barberpole. The mechanisms underlying this dynamic are unknown (lateral connections, surround suppression, feed-backs from higher layers). Here, we show that a simple mechanism such as surround-inhibition in V1 neurons can produce a significant shift in the PD of MT neurons as observed with barberpoles of different aspect ratios.