Abstract
The ubiquitous applications like Autonomous vehicles, Biometric Recognition, Security in defence sector, Medical field for disease identification, Video surveillance and Scene understanding activated vast research in the realm of Deep Learning. Deep Learning techniques [1] have emerged as a powerful tool for Feature Extraction from images or videos and also led to remarkable breach in the field of Object Detection. The field of Deep Learning [2] is predominantly becoming popular due to the improvement of Convolution Neural Network (CNN) Architectures. The Deep Learning model's accuracy depends on various Hyper-Parameters such as 'Learning Rate', 'Batch Size', 'Epoch Rate', 'Optimization Function', 'Activation Function', 'Dropout Rate' etc. Identifying the best values of these Hyper-Parameters, improves the Object Detection accuracy. This paper mainly concentrates on the selection of a better value of 'Learning Rate', at which maximum Object Detection accuracy is obtained. Different Datasets are considered for analysis and the Learning Rate at which each dataset results better accuracy is identified. After rigorous experimentation, a relationship is formulated between 'Learning Rate' and 'Dataset Size' which holds good for any dataset.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have