🧐 Bias in Image Caption Generation Systems

This repository defines an Image Caption Generation deep learning model and contains scripts to compare the sentiment of generated captions against training data.

• image-caption-generator stores the encoder, decoder and attention layers of the model based on the architecture discussed in Xu et al.'s Show, Attend and Tell: Neural Image Caption Generation with Visual Attention.
• train.py trains the model based on global hyperparameters and then produces a sample caption.
• analysis/ stores some files used to deploy Sheng et al.'s sentiment and regard classifiers on our generated captions.

Project by Jiin Kim, Arjun Kallapur, Aman Oberoi. UCLA Winter 2022. 🐻