Research Symposium

BIO

I'm AJ, I'm currently a Junior trying to get a major in Computer Science. Although I have little experience, I am interested in researching responsible artificial intelligence. In the same way, I want to dedicate myself to expanding the teaching of programming and coding to vulnerable communities and developing countries.

Fish Identifier Trained Using Synthetic Datasets

Abstract

Object detection models such as You Only Look Once (YOLO) can be trained using synthetic data. Based on previous research, other machine learning models have been successfully trained with synthetically generated data, in the case of object detection models, synthetic imagery. To train Machine learning models it is necessary to have datasets with tens of thousands of images. The generation of synthetic data allows multiplying a smaller number of real images until obtaining the amount of data necessary to train the models. In this research project, the use of the graphic software Blender is implemented, with which a dataset of synthetic images of fish is generated. After the creation of the fish dataset through Blender, YOLO will be trained in order to create a fish identifier. The expected results are that this synthetic set will provide an efficient training for the identifier. (Project in progress)

Keywords: artificial intelligence, fish, detection, identifyier, programming