A Study of Video Super-Resolution Method Using Video Coded Data as Training Data
Remina Yano, Yun Liu, Hiroshi Watanabe, Takuya Suzuki, Takeshi Chujoh, Tomohiro Ikai · 2021 IEEE 10th Global Conference on Consumer Electronics (GCCE) · 2021
This paper presents a study of video super resolution method by applying Deformable Convolution network to coded video. It is reported the effectiveness of using coded video as training data, and numerical and visual results of coded fine-tuned model. From those results, it is discussed the relationship between about characteristic of training data and especially in video’s framerate. There it is shown that video sequence which has similar framerate with training data can perform higher PSNR, since low framerate means the motion between frames is large and Deformable Convolution can learn that large motion.