Areas of Attention for Image Captioning

TLDR

Previous attention‑based image captioning models linked image regions only to the RNN state, whereas our approach directly associates caption words with image regions. We propose the Areas of Attention model, a novel attention‑based framework for automatic image captioning. The model captures dependencies among image regions, caption words, and the RNN state through three pairwise interactions, learns these associations from image‑level captions, and explores attention area generation via CNN activation grids, object proposals, and spatial transformer networks. The Areas of Attention model improves captioning by localizing relevant regions, with spatial transformer‑generated attention areas yielding the best performance and achieving state‑of‑the‑art results on MSCOCO.

Abstract

We propose "Areas of Attention", a novel attentionbased model for automatic image captioning. Our approach models the dependencies between image regions, caption words, and the state of an RNN language model, using three pairwise interactions. In contrast to previous attentionbased approaches that associate image regions only to the RNN state, our method allows a direct association between caption words and image regions. During training these associations are inferred from image-level captions, akin to weakly-supervised object detector training. These associations help to improve captioning by localizing the corresponding regions during testing. We also propose and compare different ways of generating attention areas: CNN activation grids, object proposals, and spatial transformers nets applied in a convolutional fashion. Spatial transformers give the best results. They allow for image specific attention areas, and can be trained jointly with the rest of the network. Our attention mechanism and spatial transformer attention areas together yield state-of-the-art results on the MSCOCO dataset.