Computer Vision

I'm advised by Professor Jia Deng and Erich Liang in the Princeton Vision & Learning Lab.

Github: https://github.com/Michel-Liao

You can find past computer vision posts here!

Computer Vision Posts

Notes¹

^{1. [All Coursera slides are from DeepLearning.AI]↩}

Cool External Resources

Computer Graphics

• Ray Tracing Essentials by NVIDIA

Neural Network Basics

• Neural Networks by 3Blue1Brown

PyTorch DataLoaders

• PyTorch DataLoaders and Datasets tutorial by Vihar Kurama
• Writing Custom Datasets, DataLoaders and Transforms by Sasank Chilamkurthy

Python Fundamentals

• Clearing the confusion once and for all: fig, ax = plt.subplots() by Bex T.

Projects²

• RNN and LSTM forward propagation from scratch
• Neural Style Transfer
  • Used pre-trained VGG-19 to code NST
  • Coded content and style cost function from scratch
• Face Recognition
  • Used pre-trained FaceNet for face recognition and verification
  • Coded triplet loss from scratch
• Image Segmentation
  • Used TensorFlow to code a mini-U-Net for image segmentation with 85% accuracy (sparse categorical crossentropy)
• Car Detection
  • Used TensorFlow to code non-max suppression and intersection over union to detect cars in images with pre-trained YOLO V2
• Alpaca Classifier
  • Employed transfer learning on MobileNet V2 to classify alpacas with 95% accuracy
• Sign Language Multiclass Classification
  • Used TensorFlow to build a RNN that classifies 10 signed numbers with 96% accuracy
• Cats and Dogs Classifier
  • Used PyTorch to build a modified verison of AlexNet to classify images of cats and dogs
• Sign Language Multiclass Classification
  • Used TensorFlow to build a CNN that classifies 6 signed numbers with 83% accuracy
• Facial Expression Classification
  • Used TensorFlow to build a CNN that detects a smiling face with 93% accuracy
• Convolutional Neural Network from Scratch
  • Coded padding, convolution, pooling, and backpropagation from scratch
• Optimization Methods
  • Coded mini-batch gradient descent and stochastic gradient descent with momentum, RMSprop, Adam and fixed and scheduled learning rate decay from scratch
• 4-Layer Dense Neural Network
  • Coded a 4-layer dense neural network from scratch to classify cat images with 82% accuracy
• 2-Layer Dense Neural Network for Classification
  • Coded a 2-layer neural network from scratch to classify planar data with 90% accuracy
• Handwriting Perceptron Algorithm
  • Coded MLP algorithm with 85% classification accuracy in Java
• Multiple Linear Regression for Housing Prices
  • Coded multiple linear regression from scratch to predict housing prices
• Logistic Regression for Binary Classification
  • Coded logistic regression from scratch to classify cat pictures with 70% accuracy
• Cross-correlation
  • Coded cross-correlation from scratch on any image with any filter

^{2. * indicates ongoing project. All projects done in Python unless otherwise stated.↩}

Want my computer vision posts in your inbox? Subscribe to my newsletter!