Semi-Supervised Cross-Modal Retrieval With Label Prediction [Soma Biswas, EE]

Cross-modal retrieval with image-text, sketch-image, etc. are gaining increasing importance due to abundance of data from multiple modalities, and applications like e-commerce, security, etc. In [1], we develop a novel approach which utilizes few labeled and remaining unlabeled data for this task, thus reducing manual intervention significantly. In [2], we address the problem of retrieving previously unseen data [2], since new categories are discovered dynamically in real-world.

Faculty Member: Soma Biswas [EE]

[1] D. Mandal, P. Rao, S. Biswas. Semi-Supervised Cross-Modal Retrieval With Label Prediction, IEEE Transactions on Multimedia (TMM), September, 2020.

[2] T. Dutta, A. Singh, S. Biswas. Adaptive Margin Diversity Regularizer for handling Data Imbalance in Zero-Shot SBIR, European Conference on Computer Vision (ECCV), August, 2020.

Click image to view enlarged version

Scroll Up