CNN-Projects

Implement Convolutional Neural Network Projects in TensorFlow

Handwritten Digit Recognition

• Build a CNN model and train it on MNIST dataset
• Visualize the filter in different layers
• Visualize the low-level and high-level features

Object Recognition

• Training with a Small Amount of Data (Fashion MNIST)
• One-Shot / Few-Shot / Low-Shot Learning (Cifar-100)

Image Generation

• Ability to handle large-scale human face data (CelebA) with deep neural network
• Learn and implement well-known image generation models
• Gain experience of adversarial training
• Supervised/unsupervised feature disentanglement

Action Recognition

• Ability to extract state-of-the-art deep CNN features
• Implement Recurrent Neural Networks (RNN) for action recognition
• Extend RNN models for solving sequence-to-sequence problems