RobustVisualReasoningviaLanguageGuided NeuralModuleNetworks AnonymousAuthor(s) Affiliation Address email

In the original IEP-Ref implementation, there are total 60 distinct modules in IEP-Ref. We employ early stopping based on33 validation set accuracy. While reporting accuracies on S-Ref test split, we use the model trained34 onS-Reftrainsplit. EAdditionalResults48 In this section, we provide more results comparing the performance of our model with baselines.49 Specifically, we analyze the model's performance in terms of filtering the objects based on the50 attributes color, size, shape, material, ordinality, and visibility. Results showthat ourapproach significantly outperforms baselines.