SwAMP: Swapped Assignment of Multi-Modal Pairs for Cross-Modal Retrieval
This work addresses cross-modal retrieval for applications like video and image search, but it is incremental as it builds on existing contrastive learning methods.
The paper tackles cross-modal retrieval by addressing the flawed assumption in contrastive learning that instances from different pairs are irrelevant, proposing a novel loss function based on self-labeling and swapped pseudo labels. The method achieves significant performance improvements over contrastive learning on tasks like text-based video retrieval, sketch-based image retrieval, and image-text retrieval, though specific numbers are not provided.
We tackle the cross-modal retrieval problem, where learning is only supervised by relevant multi-modal pairs in the data. Although the contrastive learning is the most popular approach for this task, it makes potentially wrong assumption that the instances in different pairs are automatically irrelevant. To address the issue, we propose a novel loss function that is based on self-labeling of the unknown semantic classes. Specifically, we aim to predict class labels of the data instances in each modality, and assign those labels to the corresponding instances in the other modality (i.e., swapping the pseudo labels). With these swapped labels, we learn the data embedding for each modality using the supervised cross-entropy loss. This way, cross-modal instances from different pairs that are semantically related can be aligned to each other by the class predictor. We tested our approach on several real-world cross-modal retrieval problems, including text-based video retrieval, sketch-based image retrieval, and image-text retrieval. For all these tasks our method achieves significant performance improvement over the contrastive learning.