Visual Turing Test

The objective of this site is to provide a common benchmark for evaluating a wide range of computer vision tasks, from 2D image-based classification and recognition, 3D scene reconstruction and registration, cross-camera tracking, to spatial and temporal reasoning about scenes and events. The goal is to promote interpretable representations and explicit bottom-up and top-down inference, in contrast to the implicit representations and task-specific training in traditional machine learning approaches. The ultimate goal is to build a framework for overcoming task traps and advancing general AI systems.