Convolutional neural network with TensorFlow / Keras (Python)
Definition and implementation of a convolutional neural network (CNN) for image classification using the TensorFlow/Keras deep learning libraries. Parallel comparison of the exact same processing steps performed using PyTorch. Example dataset: Pokémon figures. Import and preparation of the labeled training dataset. CNN architecture: convolutional processing layers vs. perceptron layers. Selection of the loss function (crossentropy) and the optimization algorithm (ADAM). Organization and execution of model training. Application of the model to a test dataset. Performance measurement.
Notebook and Data: http://tutoriels-data-science.blogspo...
Video with PyTorch: • Réseau de neurones convolutifs avec PyTorc...
Course Page: https://cours-machine-learning.blogsp...
00:00 TensorFlow - Keras for Deep Learning
04:36 Ranking "Pokémon" Figures
08:44 Starting the Notebook - Checking Versions
09:39 Loading, Inspecting, and Displaying an Image
12:00 Parsing Labeled Images from a Folder (TRAIN)
14:37 Accessing Images and Labels in the Iterator
19:50 Defining the Architecture of a Convolutional Neural Network
26:05 Loss Function and Algorithm Optimization
27:50 Training the model with fit()
29:18 Ranking on an additional image
30:36 Parsing the TEST images
31:56 Prediction in the test and calculation of accuracy