Faster R-CNN Implementation using CUDA Architecture in GeForce GTX 10 Series

Open-source deep learning tools has been distributed numerously and has gain popularity in the past decades. Training a large dataset in a deep neural network is a process which consumes a large amount of time. Recently, the knowledge of deep learning has been expand with introducing the integration between neural network and the use of graphical processing unit (GPU) which was formerly and commonly known to be used with a central processing unit (CPU). This has been one of the big leap forward in deep learning as it increases the speed of computing from weeks to hours. This paper aims to study the various state-of-the-art GPU in deep learning which included Matrix Laboratory (MATLAB) with Caffe network. The benchmark of the performance is run on three latest series of GPU platforms as of year 2017 by implementing Faster Region-based Convolutional Neural Network (R-CNN) method. Different parameters are varied to analyze the performance of mean average precision (mAP) on these different GPU platforms. The best result obtained in this paper is 60.3% of mAP using the GTX 1080.