Static Correlative Filter Based Convolutional Neural Network for Visual Question Answering
Lijun Chen, Qinyu Li, Hanli Wang, Yu Long · 2018
Visual Question Answering (VQA) has received increasing attentions due to the success of computer vision and natural language processing. The computer is required to understand the image, comprehend and reply to the question. The data modal of images makes it harder to answer than textual questions. In general, as VQA tasks use Convolutional Neural Networks (CNN) to extract image features, a better CNN model is preferred for obtaining better image representations. In this paper, the Static Correlative Filter (SCF) which is an advanced technique in convolutional layers is employed for VQA, as convolutional layer is the major component of CNN. The effectiveness of SCF for VQA is demonstrated by the experiments on the benchmark dataset of COCO-QA with two baseline image question answering models.