C 4.3 | Convolutional Neural Network Architecture | CNN Architecture | Object Detection | EvODN скачать в хорошем качестве

C 4.3 | Convolutional Neural Network Architecture | CNN Architecture | Object Detection | EvODN

Pooling: We will first understand what is Pooling. Pooling conceptually amounts to selecting the important features from a feature vector. This feature vector is usually a matrix, that is the result of Convolution. In Pooling, we usually pick the highest values from a mxn region of the feature vector in case of MaxPool or average them out in case of Average Pool. After pooling, the size of the feature vector reduces. In a CNN, what is usually done is to first apply convolution on top of the image, so as to keep the size of input and output matrix to be the same followed by a Max Pool layer to reduce the size of the matrix. This combination of Convolution and Pooling is considered as 1 layer in a CNN. And CNNs usually have many such layers cascaded one after the other, finally ending up with a smaller matrix of size mxnxp. This serves as the Feature Extractor. In case of Classical CVs, HOG and SIFT served as the feature extractors. The output matrix is then flattened to a 1D Vector. We then apply 1D convolution to this finally getting the output size that we desire. That is, if there are 20 classes in our dataset, the output feature vector should be of size 20. Even these layers can be cascaded. These layers serve as the Classifier. In case of Classical CV, SVM was used as the classifier. Finally, for ease of analysis, we convert these scores to probabilities usinig Softmax. ------------------------ This is a part of the course 'Evolution of Object Detection Networks'. See full playlist here:    • Evolution Of Object Detection Networks   ------------------------ Copyright Disclaimer: Under section 107 of the Copyright Act 1976, allowance is made for “fair use” for purposes such as criticism, comment, news reporting, teaching, scholarship, education and research.