【摘要】 为定量评价使用生成对抗网络生成的恒星和星系图像的质量,提出了感知损失函数与Mode score相结合的双样本评价方法。通过对几种常用的生成图像定量评价方法的优缺点的分析以及对感知损失函数进行的改进,得到了能够定量评价生成图像多样性和视觉质量的双样本评价模型。为了验证提出方法的有效性,分别采用Inception score,Kernal MMD,Wasserstein distance及双样本评价方法对生成的两种分辨率的恒星和星系图像进行评价,并将评价结果进行了对比分析。结果表明双样本评价方法能够全面且客观的评价生成的恒星和星系图像,对利用生成的天文图像预测未知的恒星和星系具有重要意义。更多还原

【Abstract】 In order to quantitatively evaluate the quality of the images of stars and galaxies generated by generative adversarial neural network, a two-sample evaluation method combining the perceptual loss function and Mode score was proposed. By analyzing the advantages and disadvantages of several methods commonly used for quantitative evaluation of generated images and improving the perceptual loss function, a two-sample evaluation model for evaluating the diversity and visual quality of generated images was obtained. In order to verify the effectiveness of the proposed method, the Inception score, Kernal MMD, Wasserstein distance and the method proposed in this paper, were used to evaluate the generated star and galaxy images with two resolutions, and the evaluation results were compared and analyzed. The results show that the two-sample evaluation method can comprehensively and objectively evaluate the generated images of stars and galaxies, which is of great significance for the prediction of unknown galaxies and stars.更多还原