A Proposed Exploratory Study of Object Detectors to Learn the Influence of Datasets on Model Performance

The quality of the images used to train the models in the field of object detection using deep learning models is critical in determining the model's quality. However, there are very few methods for exploring these images in datasets to see what aspects in these images have a significant impact on the model's performance. This could be one of the reasons why the models don't match human perceptions. There is a need for more study that can suggest unique methodologies to address the topic at hand because the existing literature overlooks this line of thought. As a result, this paper provides a methodology based on exploratory sequential design, which may be used to identify several aspects of images in the dataset that influence model performance.