Data augmentation for computer vision problems

• 主要作者: Wu, Rongxi
• 其他作者: Kwoh Chee Keong
• 格式: Final Year Project
• 語言: English
• 出版: Nanyang Technological University 2024
• 主題: Computer and Information Science; Data augmentation
• 在線閱讀: https://hdl.handle.net/10356/175411
• 機構: Nanyang Technological University

Computer Vision is a vital sub-field of artificial intelligence, it consists of several sub-topics such as image classification, image segmentation, object detection[1], etc. Computer vision presents a myriad of challenges that must be overcome through dedicated research and innovation. Like every research topic, it demands rigorous exploration to tackle these obstacles. In the era where data is gold, the insufficient volume of data during training would result in over-fitting, a phenomenon where a model performs exceptionally well on the training data but fails to generalize effectively to unseen data points during validation or testing. Traditionally, the size of the data set can be manually increased through the collection of new data such as by taking more pictures. However, that will require ample cost and effort. To elevate this problem, data augmentation is being employed, it involves applying transformations to existing data to generate additional examples while preserving their labels [2]. To date, many traditional data augmentation techniques are already being widely used and explored, and there is an emergence of new data augmentation techniques, that involve the generation of synthetic data points. Hence, this paper will particularly evaluate the effectiveness of training better models by using these new data augmentation techniques and the conventional transformation data augmentation techniques during data preparation, there is also an aim to address the limitations by making changes to improve the situation