Proposal of an improved loss function considering image-edge structure for DNN-based video prediction
Hiroki Nishimura, Shun-ichi Sekiguchi, Wataru Kameyama · 2024
Introduction of loss function that considers image structure to DNN-based video prediction has been proven to reduce blurriness of generated prediction frames. In this paper, we propose an improved loss function based on image gradient difference (GDL) which captures edge structure of image, and evaluate its performance over PredNet that is a well-known DNN-based video prediction scheme. Our experimental results show that the proposed loss function can improve prediction performance in terms of color representation and generating sharper prediction frames.