This dataset is a new aesthetics and attributes database (AADB) which contains aesthetic scores and meaningful attributes assigned to each image by multiple human raters. Anonymized rater identities are recorded across images allowing us to exploit intra-rater consistency using a novel sampling strategy when computing the ranking loss of training image pairs. We show the proposed sampling strategy is very effective and robust in face of subjective judgement of image aesthetics by individuals with different aesthetic tastes. Experiments demonstrate that our unified model can generate aesthetic rankings that are more consistent with human ratings. To further validate our model, we show that by simply thresholding the estimated aesthetic scores, we are able to achieve state-or-the-art classification performance on the existing AVA dataset benchmark.

Data Collection

To collect a large and varied set of photographic images, we download images from the Flickr website1 which carry a Creative Commons license and manually curate the data set to remove non-photographic images (e.g. cartoons, drawings, paintings, ads images, adult-content images, etc.). We have five different workers then independently annotate each image with an overall aesthetic score and a fixed set of eleven meaningful attributes using Amazon Mechanical Turk (AMT)2 . The AMT raters work on batches, each of which contains ten images. For each image, we average the ratings of five raters as the ground-truth aesthetic score. The number of images rated by a particular worker follows long tail distribution.