The retina is the entrance of the visual system. The flux of photons coming from the external world is converted by the retina into spike trains sent via the optic nerve and deciphered by the visual cortex in a very efficient way. Although based on common biophysical principles the dynamics of retinal neurons is quite different from their cortical counterparts, raising interesting problems for modelers. In particular most of the retinal processing is done by neuron that do not spike. Only the last step of this processing, sending the visual information to the brain, requires action potentials. This raises several questions: (1) Retina side. How does the retina structure and dynamics in response to the visual flow (especially, a moving object) shape the non stationary statistics of spike trains emitted to the brain ? What is the structure of spatio-temporal spike correlations ? What are the respective contributions of the visual flow, retinal network, non linear dynamics ? (2) Cortical side. What could be a plausible cortical model to understand how the retinal spike trains induced by a moving object are deciphered ? The answer actually depends whether one considers that statistical indicators (like firing rates) are sufficient or if individual spikes and their timing do matter. In this conference, I will address these questions introducing mathematically tractable models based on neuroscience findings, although, obviously, simplified compared to biology. I will show how dynamical and stochastic aspects can be combined to unravel (incomplete and non rigorous) interesting aspects of retinal coding/decoding of non stationary visual scene with moving objects.