Edge device-based real-time implementation of CycleGAN for the colorization of infrared video

Ruimin Huang, Huaqiang Wang, Xiaoqiao Huang, Yonghang Tai, Feiyan Cheng, Junsheng Shi

Abstract: Highlights•Our hybrid method reduced the inference time from 112 ms to 17 ms.•Lightweight models of CycleGAN were implemented.•FP32, FP16, and INT8 precision deployments of the models were implemented.•Three of our own self-collected datasets were created.•The models were deployed on three embedded hardware platforms for evaluation.