# Automatic Generation of Contrast Sets from Scene Graphs: Probing the Compositional Consistency of GQA

arXiv: https://arxiv.org/abs/2103.09591 · HTML: https://mlsanity.com/paper/2103.09591/

## Key facts

- Cited by 30 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 23 references that are in the map.
- First posted on arXiv 17 Mar 2021.

## Builds on (most cited first, 18 of 18)

- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations](https://mlsanity.com/paper/1602.07332/) (2016)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [LXMERT: Learning Cross-Modality Encoder Representations from Transformers](https://mlsanity.com/paper/1908.07490/) (2019)
- [Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference](https://mlsanity.com/paper/1902.01007/) (2019)
- [Annotation Artifacts in Natural Language Inference Data](https://mlsanity.com/paper/1803.02324/) (2018)
- [Adversarial Examples for Evaluating Reading Comprehension Systems](https://mlsanity.com/paper/1707.07328/) (2017)
- [Learning the Difference that Makes a Difference with Counterfactually-Augmented Data](https://mlsanity.com/paper/1909.12434/) (2019)
- [SWAG: A Large-Scale Adversarial Dataset for Grounded Commonsense Inference](https://mlsanity.com/paper/1808.05326/) (2018)
- [Visual7W: Grounded Question Answering in Images](https://mlsanity.com/paper/1511.03416/) (2015)
- [Evaluating Gender Bias in Machine Translation](https://mlsanity.com/paper/1906.00591/) (2019)
- [Compositional Attention Networks for Machine Reasoning](https://mlsanity.com/paper/1803.03067/) (2018)
- [Are We Modeling the Task or the Annotator? An Investigation of Annotator Bias in Natural Language Understanding Datasets](https://mlsanity.com/paper/1908.07898/) (2019)
- [Yin and Yang: Balancing and Answering Binary Visual Questions](https://mlsanity.com/paper/1511.05099/) (2015)
- [Inoculation by Fine-Tuning: A Method for Analyzing Challenge Datasets](https://mlsanity.com/paper/1904.02668/) (2019)
- [Linguistically-Informed Transformations (LIT): A Method for Automatically Generating Contrast Sets](https://mlsanity.com/paper/2010.08580/) (2020)
- [Exposing Shallow Heuristics of Relation Extraction Models with Challenge Data](https://mlsanity.com/paper/2010.03656/) (2020)
- [Latent Compositional Representations Improve Systematic Generalization in Grounded Question Answering](https://mlsanity.com/paper/2007.00266/) (2020)

## Built on by (most cited first, 30 of 30)

- [Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality](https://mlsanity.com/paper/2204.03162/) (2022)
- [VLP: A Survey on Vision-Language Pre-training](https://mlsanity.com/paper/2202.09061/) (2022)
- [VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena](https://mlsanity.com/paper/2112.07566/) (2021)
- [What You See is What You Read? Improving Text-Image Alignment Evaluation](https://mlsanity.com/paper/2305.10400/) (2023)
- [Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images](https://mlsanity.com/paper/2303.07274/) (2023)
- [Why is Winoground Hard? Investigating Failures in Visuolinguistic Compositionality](https://mlsanity.com/paper/2211.00768/) (2022)
- [The Dangers of Underclaiming: Reasons for Caution When Reporting How NLP Systems Fail](https://mlsanity.com/paper/2110.08300/) (2021)
- [Retrieval-guided Counterfactual Generation for QA](https://mlsanity.com/paper/2110.07596/) (2021)
- [ViLMA: A Zero-Shot Benchmark for Linguistic and Temporal Grounding in Video-Language Models](https://mlsanity.com/paper/2311.07022/) (2023)
- [COVR: A test-bed for Visually Grounded Compositional Generalization with real images](https://mlsanity.com/paper/2109.10613/) (2021)
- [CORE: A Retrieve-then-Edit Framework for Counterfactual Data Generation](https://mlsanity.com/paper/2210.04873/) (2022)
- [Data Efficient Masked Language Modeling for Vision and Language](https://mlsanity.com/paper/2109.02040/) (2021)
- [Rethinking Data Augmentation for Robust Visual Question Answering](https://mlsanity.com/paper/2207.08739/) (2022)
- [Measuring Compositional Consistency for Video Question Answering](https://mlsanity.com/paper/2204.07190/) (2022)
- [Break, Perturb, Build: Automatic Perturbation of Reasoning Paths Through Question Decomposition](https://mlsanity.com/paper/2107.13935/) (2021)
- [CARETS: A Consistency And Robustness Evaluative Test Suite for VQA](https://mlsanity.com/paper/2203.07613/) (2022)
- [Visual Riddles: a Commonsense and World Knowledge Challenge for Large Vision and Language Models](https://mlsanity.com/paper/2407.19474/) (2024)
- [ColorFoil: Investigating Color Blindness in Large Vision and Language Models](https://mlsanity.com/paper/2405.11685/) (2024)
- [Generalization Differences between End-to-End and Neuro-Symbolic Vision-Language Reasoning Systems](https://mlsanity.com/paper/2210.15037/) (2022)
- [The Role of Linguistic Priors in Measuring Compositional Generalization of Vision-Language Models](https://mlsanity.com/paper/2310.02777/) (2023)
- [Fighting Bias with Bias: Promoting Model Robustness by Amplifying Dataset Biases](https://mlsanity.com/paper/2305.18917/) (2023)
- [Few-Shot Dialogue Summarization via Skeleton-Assisted Prompt Transfer in Prompt Tuning](https://mlsanity.com/paper/2305.12077/) (2023)
- [Can Visual Dialogue Models Do Scorekeeping? Exploring How Dialogue Representations Incrementally Encode Shared Knowledge](https://mlsanity.com/paper/2204.06970/) (2022)
- [Exposing and Addressing Cross-Task Inconsistency in Unified Vision-Language Models](https://mlsanity.com/paper/2303.16133/) (2023)
- [Improving Data Augmentation for Robust Visual Question Answering with Effective Curriculum Learning](https://mlsanity.com/paper/2401.15646/) (2024)
- [Into the Unknown: Generating Geospatial Descriptions for New Environments](https://mlsanity.com/paper/2406.19967/) (2024)
- [HNC: Leveraging Hard Negative Captions towards Models with Fine-Grained Visual-Linguistic Comprehension Capabilities](https://mlsanity.com/paper/2605.06157/) (2026)
- [EviSafe: Evidence-Grounded Safety Evaluation for Vision-Language Models](https://mlsanity.com/paper/2608.23313/) (2026)
- [From Perception to Integration: Revisiting the Internal Dynamics of Reasoning in Vision-Language Models](https://mlsanity.com/paper/2609.34809/) (2026)
- [Evaluating Persistent Calibration under Evolving Model Knowledge](https://mlsanity.com/paper/2609.38797/) (2026)

## Abstract (arXiv)

Recent works have shown that supervised models often exploit data artifacts to achieve good test scores while their performance severely degrades on samples outside their training distribution. Contrast sets (Gardneret al., 2020) quantify this phenomenon by perturbing test samples in a minimal way such that the output label is modified. While most contrast sets were created manually, requiring intensive annotation effort, we present a novel method which leverages rich semantic input representation to automatically generate contrast sets for the visual question answering task. Our method computes the answer of perturbed questions, thus vastly reducing annotation cost and enabling thorough evaluation of models' performance on various semantic aspects (e.g., spatial or relational reasoning). We demonstrate the effectiveness of our approach on the GQA dataset and its semantic scene graph image representation. We find that, despite GQA's compositionality and carefully balanced label distribution, two high-performing models drop 13-17% in accuracy compared to the original test set. Finally, we show that our automatic perturbation can be applied to the training set to mitigate the degradation in performance, opening the door to more robust models.