vincentla / stanford-cs231n

Stanford CS 231n Convolutional Neural Networks

stanford-cs231n

Stanford CS 231n Convolutional Neural Networks

Course website: http://cs231n.stanford.edu/

Google Cloud

Go to Compute Engine on console.cloud.google.com

Note, the credentials for this project are:

Acount: Law Stanford Project Organization: law.stanford.edu

Click on the stanford-cs-231n-vm VM instance, and click "Start". Once you turn on the engine, type:

gcloud beta compute --project "stanford-cs-231n" ssh --zone "us-west1-b" "stanford-cs-231n-vm"

Jupyter Notebook

Note, that once remote, to access the Jupyter Notebook, type

jupyter notebook

in terminal as you normally would to start the jupyter notebook server. However, on the browser, go to:

http://35.212.157.230:8888/

This is based off the static IP in Google Cloud

To Push back to GitHub

Note, that to push back to GitHub, with 2FA, you get an authentication issue. For now, what I did, was generated a token (https://medium.com/@ginnyfahs/github-error-authentication-failed-from-command-line-3a545bfd0ca8). The actual token is stored as an environment variable GITHUB_TOKEN (in my local bash_profile)

Setup Instructions

For instructions on setting up Google Cloud: https://github.com/cs231n/gcloud

Setup instructions for Google Colab: https://cs231n.github.io/setup-instructions/#working-remotely-on-google-colaboratory (or can basically just google around)

Remember, if you are using Sublime Text 3 locally, you have to push changes to git and pull inside the VM to ensure that changes get passed through. Because of this, you may want to set up two terminal windows, one for Jupyter notebook and one to pull.

Lecture Notes

K Nearest Neighbor

http://cs231n.github.io/classification/

Other Public Class Resources

1. https://github.com/jariasf/CS231n
2. https://github.com/srinadhu/CS231n
3. https://github.com/Arnav0400/CS231n-2019

Recording Lectures

We can record lectures by using Quicktime Player (which comes for free on Mac). Click Record Entire Screen. Also make sure to click on Microphone to record audio.

Project Ideas

1. Sentiment Analysis: https://machinelearningmastery.com/deep-learning-bag-of-words-model-sentiment-analysis/
2. Using Yolov3 for 9000 classes: https://stackoverflow.com/questions/57853707/is-it-possible-to-use-yolo3-with-yolo9000-weights-for-more-classes
3. Yolo9000 Weights: https://awesomeopensource.com/project/philipperemy/yolo-9000

To Dos

1. 2020-05-19
   1. In the YOLOv3 notebook, move to next frame by uncommenting out a couple lines at the end of the forloop
   2. Will want to store yolo results (objects detected, confidence, and bounding boxes for each frame)
   3. Aggregate them at the video level
   4. Perform bag of words/sentiment analysis model on training data to predict sentiment
2. After (Stretch Goals)
   1. Try Yolo9000 weight for yolov3 to see if yolov3 can predict more images and if the sentiment model improves

Other Helpful Notes

1. http://cs229.stanford.edu/section/cs229-linalg.pdf