Hyperparameter Optimization for Object Detection Network

Object detection is a growing research area in the field of neural networks and deep learning with many different applications. One of the most popular applications is in the development of self-driving cars. Hyperparameter optimization is a common mechanism to improve the performance of a deep learning network. However, a lot of the research today tends to use heuristics to set the hyperparameter values. In this study, several methods are discussed and implemented to provide specific techniques that researchers can use to find optimal settings for their network. The experiments are conducted on a camera radar fusion network for detecting and classifying objects in the Nuscenes dataset scene collection. Training the model with the optimization techniques led to a 56.2% improvement in accuracy from the baseline model.