Autonet: Knowledge Graphs For Occasions Object Recognition

Object recognition in images fields is a novel and importance challenge in Computer Vision. We can through computer vision algorithms, which is the ability to acquire various knowledge about the world. Meanwhile, we are utilized knowledge to understanding about the visual world. However, the present have many state-of-the-art algorithms, which is include deep learning, but focus on using images of features to implementation applications. We have been neglected the vast amount of background knowledge of relations about the real world. In this paper, we have proposal a novel knowledge graphs framework for occasions object recognition called AutoNet, which shows using this knowledge graphs improves performance on image recognition. We utilize end-to-end learning on graphs, which is introducing the Gated Graph Neural Network (GGNN) and the Gated Graph Choose Search Neural Network (GG-CSNN) as a way of efficiently incorporating large knowledge graphs into a computer vision recognition pipeline. We have a novel method, which is a hierarchical approach for generating descriptive images paragraphs and generating a text to knowledge extraction. We have demonstrated through experiments that our method outperforms the general neural network approach through the datasets of MSCOCO-2015 and VOC2012 for images recognition.