Inter-modality Fusion based Attention for Zero-shot Cross-modal Retrieval

Due to the advances in the internet and digital image sensors, there has been a drastic in-crease in the volume of visual content. This visual content was generated and shared from various sources, such as medical archival photos, social media platforms, the education sector, and robotics. This resulted in new challenges pertaining to users’ interests, i.e., content-based image retrieval (CBIR), which is a long-established research area. As an initial step to attain the final aim of this thesis, a dive-in into content-based image retrieval using a deep convolutional network was attempted to explore the existing pooling mechanisms, diffusion mechanisms for image retrieval, and the challenges encountered in attaining the task. Experimental results are reported on various image retrieval benchmark datasets and a community archival photo dataset. Results on visual grounding, another computer vision task that has gained pace were also reported on the community archival photo dataset as a part of the initial study.