The Gracenote Multi-Crop Dataset

We present the Gracenote Multi-Crop (GNMC) dataset, to further research in algorithms for automatic image cropping. The dataset consists of a diverse collection of 10K images, each cropped in five different aspect ratios. GNMC is larger than existing datasets commonly used to benchmark image cropping approaches such as FCDB (1743 images) and FLMS (500 images), and can enable algorithms as described in "An Experience-Based Direct Generation Approach to Automatic Image Cropping" by Christensen and Vartakavi.