yirusun/stanford-cs231n-2023

Homework assignments for the Stanford CS 231n (Deep Learning for Computer Vision) class offered in Spring 2023.

CS 231n: Deep Learning for Computer Vision

This repo documents my solutions for assignments of the Stanford CS 231n (Deep Learning for Computer Vision - 2023) class. Setup details and the assignment descriptions can be found on the course website. Video lectures covering most of the course content can be found here:

• Stanford Convolutional Neural Networks for Visual Recognition (Spring 2017): Version of the class offered in 2017
• Michigan EECS 498-007 Deep Learning for Computer Visison: Covers some additional topics covered after 2017.

A brief summary of key concepts covered in different assignments is summarized below.

Assignment 1: Image Classification, kNN, SVM, Softmax, Fully Connected Neural Network

This assignment provides experience on multi-class image classification using kNN, SVM, Softmax and Fully connected neural networks using CIFAR-10 dataset. Losses and gradients for different models are derivered and are implemented using Numpy in a vectorized manner. Following are the key goals of this assignment:

• Understand the basic Image Classification pipeline and the data-driven approach (train/predict stages).
• Understand the train/val/test splits and the use of validation data for hyperparameter tuning.
• Develop proficiency in writing efficient vectorized code with numpy.
• Implement and apply a k-Nearest Neighbor (kNN) classifier.
• Implement and apply a Multiclass Support Vector Machine (SVM) classifier.
• Implement and apply a Softmax classifier.
• Implement and apply a Two layer neural network classifier.
• Understand the differences and tradeoffs between these classifiers.
• Get a basic understanding of performance improvements from using higher-level representations as opposed to raw pixels, e.g. color histograms, Histogram of Oriented Gradient (HOG) features, etc.

Assignment 2: Fully Connected and Convolutional Nets, Batch Normalization, Dropout, Pytorch & Network Visualization

Following are the key goals of this assignment:

• Understand Neural Networks and how they are arranged in layered architectures.
• Understand and be able to implement (vectorized) backpropagation for fully connected, batch norm, layer norm, dropout, among other, commonly used layers.
• Implement various update rules (SGD, RMSProp, Adam, etc.) used to optimize Neural Networks.
• Implement Batch Normalization and Layer Normalization for training deep networks.
• Implement Dropout to regularize networks.
• Understand the architecture of Convolutional Neural Networks and get practice with training them.
• Gain experience with a major deep learning framework, PyTorch.

The linked blog post was found to be very useful for understanding the backpropogarion of Batch norm and also motivated the implementation of layer norm and spatial group norm layers in this assignment.

Assignment 3: Network Visualization, Image Captioning with RNNs and Transformers, Generative Adversarial Networks, Self-Supervised Contrastive Learning

The goals of this assignment are as follows:

• Visualizing network layers and understanding key image attributes resulting in model predictions in a specific class.
  • Experimentation by making Saliency Maps, generating fooling images to increase changes of model predicting a target class and visualizing different classes.

• Understand and implement RNN, LSTM, and Transformer networks. Combine them with CNN networks for image captioning.
• Understand how to train and implement a Generative Adversarial Network (GAN) to produce images that resemble samples from a dataset.

• Understand how to leverage self-supervised learning techniques to help with image classification task by implementing SimCLR algorithm.