QualiaNet: An Experience-Before-Inference Network
For computer vision researchers, this work presents a biologically-inspired architecture for depth estimation that leverages natural scene statistics, though it is an incremental step as it only validates the concept on synthetic data without quantitative comparison to existing methods.
The paper introduces QualiaNet, a two-stage neural network that mimics human stereo vision by first computing disparity maps relative to fixation (Experience Module) and then estimating distance from disparity gradients (Inference Module). The network successfully recovers distance from disparity gradients alone, validating the proposed approach.
Human 3D vision involves two distinct stages: an Experience Module, where stereo depth is extracted relative to fixation, and an Inference Module, where this experience is interpreted to estimate 3D scene properties. Paradoxically, although our experience of stereo vision does not provide us with distance information, it does affect our inferences about visual scale. We propose the Inference Module exploits a natural scene statistic: near scenes produce vivid disparity gradients, while far scenes appear comparatively flat. QualiaNet implements this two-stage architecture computationally: disparity maps simulating human stereo experience are passed to a CNN trained to estimate distance. The network can recover distance from disparity gradients alone, validating this approach.