CUDAS: Distortion-Aware Saliency Benchmark

Visual saliency prediction remains an academic challenge due to the diversity and complexity of natural scenes as well scarcity eye movement data on where people look in images. In many practical applications, digital images are inevitably subject distortions, such those caused by acquisition, editing, compression or transmission. A great deal attention has been paid predicting distortion-free pristine images, but little given understanding impact visual distortions prediction. this paper, we first present CUDAS database - a new distortion-aware benchmark, eye-tracking was collected for 60 their corresponding 540 distorted formats. We then conduct statistical evaluation reveal behaviour state-of-the-art models provide insights building effective model The is made publicly available research community.