03 CycleGAN

For testing the CycleGAN example, I trained the model with my face and the face of the character, Neytiri, in the film Avatar. I first scrapped Neytiri images from the Bing image search using the same image scraper I made for the last DCGAN test. I also resized and center cropped all those scraped images using dataset-utils.py […]


I trained my own models for DCGAN. I first tried with the instagram pictures of one of my favorite photographer, Hamada Hideaki. I used the Instagram scrapper made by Richard Arcega for scrapping the imags. However, after finishing this first DCGAN model training, I realized that DCGAN results are hard to be satisfying with random images. Therefore, I gave it […]

01 Style Transfer

I trained my own models following the tutorial here. The paintings that I used for the training are Yoonbok Shin(1758-?)’s <A Drink Party> and Marc Chagall(1887-1985)’s <I and the Village>. The Summary of Training Process 1. Setting up Install the paperspace API npm install -g paperspace-node Login to paperspace from your terminal paperspace login Clone the repository […]

[Final] Proposal

The Character Network and Sentiment in an Epic Story For my final project of this A to Z class, I am going to analyze the importance of characters, relationships between them and sentiment in those relationships in an epic story and visualize the result. I am interested in text analyzing stuff such as word2vec, and sentiment […]

05 Critics!

This project created a word2vec matrix based on 56326 movie reviews scrapped from Rotten Tomato. I used ml5 library for generating the matrix. It shows 10 closest words to your input word according to the matrix. You can see more logical correlation between the input and outputs when you give it an adjective that is […]

04 Sound of Marcov

This is a generative music project that creates musical sequences from the soundtrack of Sound of Music. I used Tone.js to play the samples, but didn’t quite understand how to play the keys in sequence and in consequence, the webpage is a bit buggy when it plays the samples now. (to be fixed) Visit the website