Cross-modal image aesthetics quality evaluation method based on knowledge distillation

The invention relates to a cross-modal image aesthetic quality evaluation method based on knowledge distillation. The method comprises the following steps of: S1, designing a teacher network for image-text multi-modal aesthetic quality evaluation; S2, designing a student network for image aesthetics quality evaluation; S3, designing a discriminator network for adversarial training; S4, designing a loss function for training the teacher network and the student network; S5, training the networks designed in the steps S1, S2 and S3 by using the loss function; S6, inputting an image for testing into the trained student network to predict the aesthetic quality of the student network. The method provided by the invention can significantly improve the aesthetic quality prediction precision. 本发明涉及一种基于知识蒸馏的跨模态图像美学质量评价方法。包括以下步骤：S1：设计图文多模态美学质量评价的教师网络；S2：设计用于图像美学质量评价的学生网络；S3：设计用于进行对抗训练的判别器网络；S4：设计用于训练教师网络和学生网络的损失函数；S5：使用损失函数对步骤S1、S2及S3所设计网络进行训练；S6：将用于测试的图像输入训练好的学生网络预测其美学质量。本发明方法能显著提高美学质量预测精度。