data/README.md

Data requirements

To execute the code the following data are needed, once downloaded the path to the data must be specified in the misc/config.py file.

• Ms-CoCo dataset (annotations and images)

• Ms CoCo rest-val split from "Deep Visual-Semantic Alignments for Generating Image Descriptions" by Karpathy et al.

• Word embedding and dictionnary from the paper "Skip-Thought Vectors" by Kiros et al.

• Pre-initialized weights of the image pipeline

Additionnal data for localization evaluation

• Visual Genome dataset (images and data and region descriptions)