Abstract:
"
VQA has emerged as a multidisciplinary challenge that integrates both vision and language
processing. It has gained a lot of interest from both computer vision and natural language
processing communities. Simply put, given an image and a natural language question about the
image, a VQA task requires the system to find the correct answer by combining visual features of
the image with inferences drawn from the question. A successful system must be able to
comprehend an image semantically, understand the textual input, and generate a response based
on its visual, textual, and logical interpretation of the inputs. This is typically done by making use
of deep learning based techniques that extract the visual features of the image and the textual
features of the question. Many researchers are interested in VQA because of its various application.
Multiple methods and approaches have been utilized to propose a number of VQA models over
time.
The author believes that given the complexity of an image, simply using the image and the question
is insufficient for a VQA model to perform well. Considering this, the author proposes CA-VQA,
an experiment that aims to validate the use of additional textual information about an image in
order to produce better results. A third input is used by CA-VQA to define the image. The accuracy
of the proposed method is 59.19%, which is considered acceptable"
