SalGAN: Visual Saliency Prediction with Generative Adversarial Networks

TL;DR

SalGAN is a deep learning model that uses adversarial training to improve the accuracy of visual saliency prediction, achieving state-of-the-art results across multiple metrics.

Contribution

This paper introduces SalGAN, the first to apply adversarial training with GANs to enhance visual saliency prediction accuracy.

Findings

Achieves state-of-the-art performance on saliency benchmarks

Adversarial training improves prediction quality

Source code and models are publicly available

Abstract

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/.