Paper ID: 2210.03838

Learning to embed semantic similarity for joint image-text retrieval

Noam Malali, Yosi Keller

We present a deep learning approach for learning the joint semantic embeddings of images and captions in a Euclidean space, such that the semantic similarity is approximated by the L2 distances in the embedding space. For that, we introduce a metric learning scheme that utilizes multitask learning to learn the embedding of identical semantic concepts using a center loss. By introducing a differentiable quantization scheme into the end-to-end trainable network, we derive a semantic embedding of semantically similar concepts in Euclidean space. We also propose a novel metric learning formulation using an adaptive margin hinge loss, that is refined during the training phase. The proposed scheme was applied to the MS-COCO, Flicke30K and Flickr8K datasets, and was shown to compare favorably with contemporary state-of-the-art approaches.

Submitted: Oct 7, 2022