For my final project I’m interested in creating a “critiquing machine”.

In short, I’d like to have interactors upload an image, and my program will return a piece of text which is formatted in the style of a critique or art review about that image.

I’ve been considering using Char-RNN as my text generation software. As my input dataset, I have collected around 2000 art reviews from Art in America which should provide a solid base. Here’s a link to my GitHub page where I talk about my reasoning for choosing the Art In America reviews.

For my initial image set, I am considering using (by recommendation of Golan) deviantArt images. My current problem lies in caption generation as a seed for Char-RNN. Using the description of the images is too varied and may not produce effective results.