Monkeys

Developed in the Neural Networks course by: Diego Quezada and Kevin Reyes.

Objectives

• Apply convolutional neural networks to the classification of monkey breeds using Transfer Learning.
• Visualize the internal state of the convolutional network in an interpretable way using CAM.
• Compare VGG16 and VGG19 network performance.
• Compare CNN network performance, CNN with skip connections trained exclusively to classify monkeys with networks with Transfer Learning.

Description

Convolutional neural networks for the classification of monkey breeds, using skip connections, transfer learning and CAM method for the visualization of the internal state of the network.

Conclusions

• Neural networks with Transfer Learning have superior performance to networks trained exclusively for monkey classification, since they store a large set of patterns in their parameters, which can be useful for object recognition in other classification problems.
• The CAM method is a very useful tool to visualize the internal state of the convolutional network and detect errors that the network makes when predicting in an interpretable way.
• The face of the monkeys is usually a point of great attention for the network when predicting the race of the monkey.

Technology stack

• Numpy.
• Pandas.
• Matplotlib.
• Seaborn.
• Cv2.
• Keras.
• Tensorflow.
• VGG16.
• VGG19.