Resolving Referring Expressions in Images With Labeled Elements
This addresses the problem of visual-language grounding for images with textual annotations, which is incremental as it builds on existing referring expression resolution methods by incorporating labeled elements.
The paper tackles the problem of resolving natural language referring expressions in images that contain labeled elements (text with bounding boxes), presenting an end-to-end trainable architecture that embeds element information into image feature maps. The result shows an improvement over image-only methods and other element-incorporating approaches, demonstrated on COCO-based datasets and a new webpage image dataset.
Images may have elements containing text and a bounding box associated with them, for example, text identified via optical character recognition on a computer screen image, or a natural image with labeled objects. We present an end-to-end trainable architecture to incorporate the information from these elements and the image to segment/identify the part of the image a natural language expression is referring to. We calculate an embedding for each element and then project it onto the corresponding location (i.e., the associated bounding box) of the image feature map. We show that this architecture gives an improvement in resolving referring expressions, over only using the image, and other methods that incorporate the element information. We demonstrate experimental results on the referring expression datasets based on COCO, and on a webpage image referring expression dataset that we developed.