SalGAN: Visual Saliency Prediction with Generative Adversarial Networks-猫眼课题宝

DOI：

发表时间：

2017-01

期刊：

ArXiv

影响因子：

通讯作者：

中科院分区：

其他

文献类型：

作者：

Junting Pan;C. Canton-Ferrer;Kevin McGuinness;N. O’Connor;Jordi Torres;E. Sayrol;Xavier Giro-i-Nieto-Xavier-Giro-i-Nie

我们介绍了SalGAN，这是一种深度卷积神经网络，用于通过对抗性示例进行视觉显着性预测。该网络的第一阶段由一个生成器模型组成，其权重通过反向传播来学习，该反向传播是从显着图的下采样版本上的二进制交叉熵（BCE）损失计算的。由此产生的预测是由一个神经网络进行处理的，该神经网络经过训练以解决由生成阶段生成的显着性图和地面实况图之间的二进制分类任务。我们的实验展示了对抗性训练如何在与BCE等广泛使用的损失函数相结合时，在不同的指标上达到最先进的性能。我们的结果可以通过https://imatge-upc.github上的源代码和训练模型复制。io/saliency-salgan-2017/.

We introduce SalGAN, a deep convolutional neural network for visual saliency prediction trained with adversarial examples. The first stage of the network consists of a generator model whose weights are learned by back-propagation computed from a binary cross entropy (BCE) loss over downsampled versions of the saliency maps. The resulting prediction is processed by a discriminator network trained to solve a binary classification task between the saliency maps generated by the generative stage and the ground truth ones. Our experiments show how adversarial training allows reaching state-of-the-art performance across different metrics when combined with a widely-used loss function like BCE. Our results can be reproduced with the source code and trained models available at https://imatge-upc.github. io/saliency-salgan-2017/.

SalGAN: Visual Saliency Prediction with Generative Adversarial Networks

文献摘要

被引文献

相似文献