Conventional methods of object detection and image classification rely heavily on manual annotations. As we are stepping into the era of Big Data, such methods of annotation are inefficient with millions of images. So we need to design an algorithm to efficiently and autonomously do detection and classification.
This page revisits the problem of visual concept learning, taking the advantage of recent progress in deep learning, computer vision, and natural language processing, and develops a fully automatic visual concept learning algorithm using parallel text and visual corpora.
Try pictures here and see the results.
Here is the result.
Department of Electronic Engineering, Shanghai Jiao Tong University.