Abstract:
Image classification is the task of finding class of an input image or probabilities of the
classes that best describe an input image. Deep Convolutional Neural networks (CNN)
are considered nowadays a popular deep learning technique for visual recognition tasks
and may achieve performance even exceeding human performance in some areas. CNNs
are dependent on their sizes and quality of training data. They are still not robust to
visual artifacts such as glare and noise, which humans are capable to cope with.
Different Deep CNNs architectures are studied, trained and tested on DIGITS, a framework
developed by NVIDIA to train and test deep learning models. Different Architectures
namely LeNet 5, Alexnet and GoogleNet are trained on different standard datasets
(MNIST, CIFAR 10 and PLANKTON) and tested. Classification Results are analyzed
and presented. .
