Unimodal experiments, where both the query and database exist in the same feature space (e.g., images), are commonly conducted on six popular, freely available image datasets: LabelMe, CIFAR-10, NUS-WIDE, MNIST, SIFT1M, and ImageNet. These datasets vary greatly in size, ranging from 22,019 to 1.3 million images, and are represented by diverse feature descriptors like GIST, SIFT, RGB pixels, and bag-of-visual-words. The content spans a wide array of image topics, from natural scenes to personal photos, logos, and drawings, offering rich resources for unimodal hashing research.
Cross-modal retrieval experiments, where the query and database are in different feature spaces (e.g., image and text), are typically performed on the Wiki, Microsoft COCO, and NUSWIDE datasets. Each of these datasets includes images paired with textual descriptions, which are essential for training and evaluating cross-modal retrieval models.
Below is a list of key publications and their associated datasets: