CS231n: Assignment Solutions

Stanford - Spring 2025

About

Overview

Complete solution code of standford cs231n assignemnt (spring 2025). Check out a detailed walkthrough at this link. The post contains more comprehensive explanations.

Main sources

• (official) Course page
• (official) Assignments1
• (official) Assignments2
• (official) Assignments3
• Sangjae's solution blog post

Solutions

Assignment 1

• Q1: k-Nearest Neighbor classifier. (Done)
• Q2: Implement a Softmax classifier. (Done)
• Q3: Two-Layer Neural Network. (Done)
• Q4: Higher Level Representations: Image Features. (Done)
• Q5: Fully-connected Neural Network. (Done)

Assignment 2

• Q1: Batch Normalization. (Done)
• Q2: Dropout. (Done)
• Q3: Convolutional Networks. (Done)
• Q4 : PyTorch on CIFAR-10. (Done)
• Q5: Image Captioning with Vanilla RNNs (Done)

Assignment 3

• Q1: Image Captioning with Transformers (Done)
• Q2: Self-Supervised Learning for Image Classification (Done)
• Q3: Denoising Diffusion Probabilistic Models (Done)
• Q4: CLIP and Dino (Done)

Running Locally

Instead of relying on Google Colab, I’ve set things up to run on a local GPU environment for better control and performance. All necessary dependencies and environment configurations are predefined in a Dockerfile.
After chaning directory to cs231n-solution. build the docker

  docker build --tag pytorch-gpu .
  docker run --gpus=all -d -it --privileged  --name pytorch-gpu pytorch-gpu