Have you ever wondered during a night how far that dark building is away from you?

We present a method that can robustly determine the object distances from a pair of noisy images with different depth-of-fields. It analyzes the blurriness of the image boundaries to predict the object distance using depth from defocus. 

The method outperforms previous depth from defocus algorithms, both learning and physics-based. It robustly generates depth maps on images with 4x more noise than typically used in previous methods. [Read More]

Do you want a 3D camera that is portable, robust to motion, and does not emit light? 

Here is Focal Split,  a handheld, snapshot, passive depth camera with fully onboard power and computing (4.9W and 500 FLOPs per pixel). It is based on a custom optical system that mimics the eyes of jumping spiders, and the data is solely processed by a Raspberry PI. A DIY guide costing $500 is provided. [More information to come]