End-to-End Video Compressive Sensing Using Anderson-Accelerated Unrolled Networks
Yuqi Li, Miao Qi, Rahul Gulve, Mian Wei, Roman Genov, Kiriakos N. Kutulakos, Wolfgang Heidrich
Abstract
High-frame-rate imaging is concerned with recording videos at rates in excess of hundreds of frames per second. However, with bandwidth being a limiting factor, conventional cameras record either a very low spatial resolution with a relatively high frame rate or relatively high spatial resolution with a low frame rate.
Using mask-based compressive sensing, it becomes feasible to capture high-frame-rate and high spatial-resolution videos with an efficient spatio-temporal encoding. The encoding is achieved using the CEP image sensor. The exposure control can be viewed as an encoding of the captured frames with a set of binary temporal masks.
With coded-exposure cameras, it is possible to encode multiple subframes into a captured image and decode them later using frame reconstruction methods