Using Text to Teach Image Retrieval
This work addresses the problem of improving image retrieval, particularly when image data is limited, for users who need more robust representations for image retrieval based on textual instructions.
This paper proposes representing the image feature space as a graph, where neighborhoods are defined by geodesic distance. To overcome sparse sampling when images are limited, they augment the manifold with geometrically aligned text, using text to improve image retrieval. They also introduce a new public dataset based on CLEVR to quantify semantic similarity between visual and text data.
Image retrieval relies heavily on the quality of the data modeling and the distance measurement in the feature space. Building on the concept of image manifold, we first propose to represent the feature space of images, learned via neural networks, as a graph. Neighborhoods in the feature space are now defined by the geodesic distance between images, represented as graph vertices or manifold samples. When limited images are available, this manifold is sparsely sampled, making the geodesic computation and the corresponding retrieval harder. To address this, we augment the manifold samples with geometrically aligned text, thereby using a plethora of sentences to teach us about images. In addition to extensive results on standard datasets illustrating the power of text to help in image retrieval, a new public dataset based on CLEVR is introduced to quantify the semantic similarity between visual data and text data. The experimental results show that the joint embedding manifold is a robust representation, allowing it to be a better basis to perform image retrieval given only an image and a textual instruction on the desired modifications over the image