Grounded Latent Trees for Visual Question Answering
This repository contains the code associated with the paper Latent Compositional Representations Improve Systematic Generalization in Grounded Question Answering, published in TACL (2020).
Setup
- Install all requirements
pip install -r requirements.txt
Arithmetic Dataset
- Run with this AllenNLP command:
allennlp train train_configs/glt_arithmetic.jsonnet -s experiments/experiment_name --include-package src
CLEVR / CLOSURE Dataset
-
Download CLEVR image features + preprocessed data (image features are extracted with py-bottom-up-attention) into the
data
directory. -
Run with this AllenNLP command:
allennlp train train_configs/glt_clevr.jsonnet -s experiments/experiment_name --include-package src