Channel: Learn By Watch

Category: Education

Tags: cnnconvolutional neural network tutorialcnn machine learningconvolutional neural networkmachine learning coursepadding in convolutional neural networkcnn neural networkcnn machine learning algorithmconvolutional neural network pythoncnn metrics architecturecnn machineconvolutional neural network theorem problemsobject detection using convolutional neural networkcnn architectureimage classification convolutional neural network

Description: A Convolutional Neural Network (ConvNet/CNN) is a Deep Learning algorithm which can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other. We will start with convolutional neural networks and it is mainly used in computer vision problems. ● Types of computer vision problems were discussed: ○ Image classification ○ Image classification with localisation ○ Object detection ● Why cannot we use normal neural networks for solving these kinds of problems? We looked at 2 important issues: ○ Problems with high quality images - We saw that with high definition images the number of parameters to train increases drastically and thus increases our training time a lot. ○ Problems with modified images - Normal neural networks cannot tell the similarity between two images of the same object but at different locations. ● Convolution step - We looked at the very basic step of convolution and by taking an example, did one convolution. ● Padding - Padding is added so that we can make our input and output matrices of the same size. ● Types of padding : ○ Valid padding - Where no padding is added. ○ Same padding - Padding is added such that the size of the input and the output matrices are of the same size. ● Strides - The number of steps to be taken by the filter after each convolution step. ● Dimension of the output matrix - We combined padding and strides to look at the overall dimension of the output layer.