GRACE: Designing Generative Face Video Codec via Agile Hardware-Centric Workflow

The recent paper on the Animation-based Generative Codec (AGC) presents a groundbreaking deployment scheme tailored for FPGA platforms, specifically designed to tackle the challenges of video compression on edge devices. These challenges include high power consumption and the complexity of adapting to evolving algorithms. By employing advanced network compression methods and a co-processor paradigm, the authors have developed an AGC FPGA prototype on the PYNQ-Z1 platform. This prototype demonstrates remarkable energy efficiency, achieving 24.9 times higher efficiency compared to traditional CPUs and 4.1 times higher than GPUs, while only consuming 11.7 microjoules per pixel for reconstruction. This advancement is crucial for enhancing video processing capabilities in environments where resources are limited, paving the way for more efficient and adaptable video services.