Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning

TL;DR

This paper introduces an adaptive attention model with a visual sentinel that dynamically decides when to attend to image regions or rely on language context, improving image captioning performance.

Contribution

The paper proposes a novel adaptive attention mechanism with a visual sentinel, enabling selective focus on image regions or language context during caption generation.

Findings

Achieves new state-of-the-art results on COCO and Flickr30K datasets.

Significantly improves captioning accuracy by adaptive attention.

Reduces unnecessary visual attention for non-visual words.

Abstract

Attention-based neural encoder-decoder frameworks have been widely adopted for image captioning. Most methods force visual attention to be active for every generated word. However, the decoder likely requires little to no visual information from the image to predict non-visual words such as "the" and "of". Other words that may seem visual can often be predicted reliably just from the language model e.g., "sign" after "behind a red stop" or "phone" following "talking on a cell". In this paper, we propose a novel adaptive attention model with a visual sentinel. At each time step, our model decides whether to attend to the image (and if so, to which regions) or to the visual sentinel. The model decides whether to attend to the image and where, in order to extract meaningful information for sequential word generation. We test our method on the COCO image captioning 2015 challenge dataset and Flickr30K. Our approach sets the new state-of-the-art by a significant margin.