E$^2$SR: An End-to-End Video CODEC Assisted System for Super Resolution Acceleration
TimeTuesday, July 12th2:37pm - 3pm PDT
Location3002, Level 3
Event Type
Research Manuscript
AI/ML Design: Circuits and Architecture
DescriptionIn this paper, we explore an end-to-end video CODEC assisted hardware software co-design solution for the SR task, which for the first time tightly couples the working principle of the cloud server with the device to provide a smooth high-definition video viewing experience. To alleviate the per-frame DNN processing in the device, we search the motion vectors in the cloud server and then pack them as addons into the bitstream waiting for future HR frames reconstruction. We also offer the corresponding microarchitecture design to enable the reconstruction algorithm in the device.