Semantic Compositional Networks for Visual Captioning
This addresses caption generation for images and videos, with incremental improvements in performance.
The paper tackles image captioning by developing a Semantic Compositional Network (SCN) that uses detected semantic tags to compose LSTM parameters, extending it to video captioning. Experimental results on COCO, Flickr30k, and Youtube2Text datasets show it significantly outperforms prior state-of-the-art approaches across multiple metrics.
A Semantic Compositional Network (SCN) is developed for image captioning, in which semantic concepts (i.e., tags) are detected from the image, and the probability of each tag is used to compose the parameters in a long short-term memory (LSTM) network. The SCN extends each weight matrix of the LSTM to an ensemble of tag-dependent weight matrices. The degree to which each member of the ensemble is used to generate an image caption is tied to the image-dependent probability of the corresponding tag. In addition to captioning images, we also extend the SCN to generate captions for video clips. We qualitatively analyze semantic composition in SCNs, and quantitatively evaluate the algorithm on three benchmark datasets: COCO, Flickr30k, and Youtube2Text. Experimental results show that the proposed method significantly outperforms prior state-of-the-art approaches, across multiple evaluation metrics.