Embracing New Techniques in Deep Learning for Estimating Image Memorability

Various works have suggested that the memorability of an image is consistent across people, and thus can be treated as intrinsic property image. Using computer vision models, we make specific predictions about what people will remember or forget. While older work has used now-outdated deep learning architectures rooted in shallow visual processing to predict memorability, innovations field given us new techniques apply this problem. Here, propose evaluate five alternative models which exploit developments from last 5 years, largely introduction residual neural networks, are intended allow model use semantic information estimation process. These were tested against prior state art with a combined dataset built optimize both within-category across-category predictions. Our findings suggest key network had overstated its generalizability was overfit on training set. outperform model, leading conclude networks simpler convolutional regression. We our state-of-the-art readily available research community, allowing memory researchers wider range images.