In this work, we show an application of discriminative localization [20] that allows users to search for images with multiple spatial and conceptual constraints by combining a discriminatively trained deep CNN ending in an average pool with a joint-embedding language model [11] and a custom database engine. We apply ...
