Repository for code and data used in "Probing Cross-Modal Representations in Multi-Step Relational Reasoning" by Parfenova, Elliott, Fernández and Pezzelle (2021).
Images from the 3POS1 dataset can be downloaded here: 3POS1 Images.
Faster R-CNN features for 3POS1 images, needed for fine-tuning pre-trained LXMERT are available here: 3POS1 Image Features (~7 GB).