Processing pipeline. A, Acquisition geometry for the Hunter–Hibbard database. The setup mimicked the typical geometry of the human visual system (see more details in the text). This geometry is crucial for capturing realistic disparity statistics from natural scenes. A stereoscopic reconstruction (red-cyan anaglyph) of a sample from the dataset is also shown. B, Input images captured by the left and the right camera for the scene shown in A. The subtle differences in the images (disparity) provide important cues about the 3-D structure of the scene. Depending on the simulation, an ROI in the visual field was identified, and inputs were sampled only within this ROI (foveal ROI illustrated in purple). Each sample consisted of two patches with equal retinal coordinates in the left and the right eye (illustrated in red and green, respectively). All simulations presented in this article were run using N = 100,000 samples each. C, The processing pipeline. The samples were processed sequentially, one per iteration. The processing of the ith sample is shown in the box. First, the ON/OFF processing in the LGN was modeled using difference-of-Gaussian filters in the two eyes—leading to left-ON, left-OFF, right-ON, and right-OFF maps. The activations of the four maps were then converted to first-spike relative latencies using a simple inverse operation (first-spike latencies for only five neurons per LGN map shown in the figure). After thresholding, the earliest 10% of first spikes were allowed to propagate to the V1 layer. The V1 layer consisted of 300 neurons connected to the LGN maps through Hebbian synapses endowed with STDP. A winner-take-all inhibition scheme was also implemented to ensure no more than one neuron fired per iteration. The first spikes from the ith sample were propagated through the synapses, and this was followed by a change in synaptic weights dictated by the STDP algorithm (see text for details about the implemented STDP model). This concluded the processing of the ith sample and was followed by the (i + 1)^th sample, which was processed in a similar fashion.