Context-Adaptive-based Image Captioning by Bi-CARU

Image captions are abstract expressions of content representations using text sentences, helping readers to better understand and analyse information between different media. With the advantage encoder-decoder neural networks, can provide a rational structure for tasks such as image coding caption prediction. This work introduces Convolutional Neural Network (CNN) Bidirectional Content-Adaptive Recurrent Unit (Bi-CARU) (CNN-to-Bi-CARU) model that performs bidirectional consider contextual features captures major feature from image. The encoded coded form is respectively passed into forward backward layer CARU refine word prediction, providing output captioning. An attention also introduced collect produced by context-adaptive gate in CARU, aiming compute weighting relationship extraction determination. In experiments, proposed CNN-to-Bi-CARU outperforms other advanced models field, achieving detailed representation captions. obtains score 41.28 on BLEU@4, 31.23 METEOR, 61.07 ROUGE-L, 133.20 CIDEr-D, making it competitive captioning MSCOCO dataset.