Exploration Study of Ensembled Object Detection Models and Hyperparameter Optimization

Object identification models are becoming more accurate as processing capabilities improve. It is our goal to improve the accuracy of object recognition by the use of several ensembles of distinct state-of-the-art object detection models. The use of single architectures and models to handle object detection challenges has been demonstrated in prior studies; however, each model was later shown to have its own bias and variation. "Ensemble Learning" is currently being studied in recent research after the success of fundamental ensembled models like XGBoost. Ensemble learning in object detection is proposed to be expanded through this research by grouping different permutations of existing models to reduce individual bias and variance while improving metrics, accuracy, and gathering metrics that will aid in hyperparameter optimization for future research on object detection ensembles. It took us a while to find a top-performing ensemble for PASCAL VOC problems.