Master in Computer Vision (UAB) - M3 Machine Learning

Implementation of machine learning and deep learning techniques for image classification.

The code of the different methods is organised in directories.

• Session 1: SVM classifier: A chosen descriptor (SIFT) computes the features of the images that are classified by a SVM.
• Session 2: Bag Of Visual Words
• Session 3: Fisher Vectors
• Session 4: CNN features + SVM: Two approaches considered:
• Using the already trained VGG model, the features of the last fully connected layer are provided to the SVM classifier.
• Use the Bag Of Visual Words approach taking as input the features from an inner layer of the VGG model.
• Session 5: Fine-tune CNN:
• Change the last fully connected layer of the already trained VGG model to match the number of classes of our dataset.
• Take the output of the previous convolutional layer of the VGG model and add fully connected layers, getting a more compact network.
• Session 6: CNN from scratch: Train a proposed CNN model from scratch.

• Idoia Ruiz
• Roque Rodriguez
• Lidia Talavera
• Onofre Martorell