Image classification algorithms such as Convolutional Neural Network used for classifying huge image datasets takes a lot of time to perform convolution operations, thus increasing the computational demand of image processing. Compared to CPU, Graphics Processing Unit (GPU) is a good way to accelerate the processing of the images. Parallelizing multiple CPU cores is also another way to process the images faster. Increasing the system memory (RAM) can also decrease the computational time of image processing. Comparing the architecture of CPU and GPU, the former consists of a few cores optimized for sequential processing whereas the later has thousands of relatively simple cores clocked at approx. 1Ghz. The aim of this project is to compare the performance of parallelized CPUs and a GPU. Python’s Ray library is being used to parallelize multicore CPUs. The benchmark image classification algorithm used in this project is Convolutional Neural Network. The dataset used in this project is Plant Disease Image Dataset. Our results show that the GPU implementation achieves 80% speedup compared to the CPU implementation
It has always been cumbersome to process real time images. Studies showed that there can be two ways to analyse this. One hand is about central processing unit (CPU) and the other is about Graphics Processing Unit (GPU). To obtain highest possible performance they have to be used at the same time. This project will compare the performance of CPU and GPU for real time image processing. The main drawback of Python’s Multiprocessing module is that it cannot be used for handling large numeric data. It cannot be used in Deep Learning Frameworks such as Keras as it decreases the accuracy of the models. In recent years, parallel computing and soft computing has become a rapidly evolving field of study. The demand for parallel processing in increasing day by day. There are various software tools and libraries by which we can parallelize our programs. For example, we have OPENMP in c++ for parallel computing. OPENMP supports FORTRAN, C and C++.
It is basically an Application Programming Interface for shared Memory Model programming. Python has its separate parallel processing module named Multiprocessing. Multiprocessing module enables to spawn multiple processes, allowing programmer to fully leverage the computing power of multiple processors. The main drawback of Python’s Multiprocessing module is that it cannot be used for handling large numeric data. It cannot be used in Deep Learning Frameworks such as Keras as it decreases the accuracy of the models. Shared variables cannot be used in the Multiprocessing Module. Python also has a Parallel and Distributed computing framework called Ray. Ray can be used for developing emerging AI applications such as image classification, face recognition etc. Parallelizing multiple cores of CPU using Ray can also increase the speedup of the model significantly. The benchmark image classification algorithm used in this project is Convolutional Neural Network. The dataset used in this project is Plant 4 Disease Image dataset containing around 30000 images. The system is configured with 16 GB RAM with 4 CPU Cores and Tesla P100 GPU. This project compares the performance of 2-core, 3-core and 4-core parallelized CPUs with GPU.
Dataset Link: https://www.kaggle.com/emmarex/plantdisease