Image captioning using encoder-decoder neural network architectures
Loading...
Date
2023-10-09
Authors
Ζαραδούκας, Ευστράτιος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In the ever-evolving landscape of the digital age, harnessing vast volumes of data, especially
in the visual realm, stands as a paramount challenge. The thesis at hand explores the fascinating
domain of ’Image Captioning’, an intersection of Computer Vision and Natural Language
Processing tasked with translating image content into cohesive, human-readable descriptions.
Anchored on Neural Networks, the backbone of contemporary Artificial Intelligence, this study
delves into foundational structures like Multilayer Feedforward Neural Networks and more advanced
architectures like Convolutional Neural Networks, Recurrent Neural Networks, and the
groundbreaking Transformers.
Drawing from previous academic research, our discussion is set against the backdrop of historical
changes in models and the latest methods. Empirically, our study takes root in acclaimed
datasets: the Flickr8k and the MSCOCO, benchmarked against commonly-used metrics defining
the quality of image captioning models. This thesis fuses theory with hands-on experiments,
offering insights into the prowess of various neural network configurations for image captioning,
particularly when constrained by limited resources. In summary, we explain the basics, review
past research, describe the datasets we used, explain how we built our methods, show our tests
and results, and end with our thoughts, conclusions, and ideas for future work. Through this
study, we aim to develop an image captioning system, capable enough to show the potential and
issues of using neural networks in the area of image captioning.
Description
Keywords
Machine learning, Neural networks, Deep learning, Computer vision, Natural language processing, Image captioning