Performance Analysis on Fine-tuned Region-based CNN for Object Recognition

Abstract

Abstract—In this time, machine learning technology has developed rapidly, especially on deep learning architecture where many models have been created and already has a good result. However, to create a good system it took a long process such as designing a model architecture, creating a vast number of dataset and test the model many times to obtain the best performance. It is practically hard to create such a system. Therefore ï¬ne-tuning is a fascinating thing to discuss, by merely taking a model that has been known having a good result and conï¬gure the model to suit our needs, ï¬ne-tuning becomes more popular method than to create a model from scratch. In this experiment, we tried to ï¬ne-tune the R-CNN model where the pre-trained model used the Residual network architecture. Our best is at 90%. Index Terms—Deep learning, Neural network, Fine-tuning