Underwater Image Super-Resolution using Deep Residual Multipliers

摘要: We present a deep residual network-based generative model for single image super-resolution (SISR) of underwater imagery use by autonomous robots. also provide an adversarial training pipeline learning SISR from paired data. In order to supervise the training, we formulate objective function that evaluates \textit{perceptual quality} based on its global content, color, and local style information. Additionally, USR-248, large-scale dataset three sets images 'high' (640x480) 'low' (80x60, 160x120, 320x240) spatial resolution. USR-248 contains instances supervised 2x, 4x, or 8x models. Furthermore, validate effectiveness our proposed through qualitative quantitative experiments compare results with several state-of-the-art models' performances. analyze practical feasibility applications such as scene understanding attention modeling in noisy visual conditions.