mlsanityData to 14 Oct 2026

Explanation-Driven Counterfactual Testing for Faithfulness in Vision-Language Model Explanations

arXiv 2510.00047 ↗First posted 27 Sep 2025cs.CV cs.AIcited by 12

On the map: Multimodal LLMs / vision-language models → VLM benchmarks and visual reasoning

Abstract (arXiv)

Vision-Language Models (VLMs) often produce fluent Natural Language Explanations (NLEs) that sound convincing but may not reflect the causal factors driving predictions. This mismatch of plausibility and faithfulness poses technical and governance risks. We introduce Explanation-Driven Counterfactual Testing (EDCT), a fully automated verification procedure for a target VLM that treats the model's own explanation as a falsifiable hypothesis. Given an image-question pair, EDCT: (1) obtains the model's answer and NLE, (2) parses the NLE into testable visual concepts, (3) generates targeted counterfactual edits via generative inpainting, and (4) computes a Counterfactual Consistency Score (CCS) using LLM-assisted analysis of changes in both answers and explanations. Across 120 curated OK-VQA examples and multiple VLMs, EDCT uncovers substantial faithfulness gaps and provides regulator-aligned audit artifacts indicating when cited concepts fail causal tests.

Key facts. Cited by 12 papers in the map (data through 14 Oct 2026). Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 17 of its 23 references that are in the map. First posted on arXiv 27 Sep 2025.

Builds on

17 of 23 references.

References · 17

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  3. 2025Qwen2.5-VL Technical Reportcited by 5,587
  4. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,796
  5. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  6. 2023InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Taskscited by 2,565
  7. 2021SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equationscited by 1,937
  8. 2024A Survey on LLM-as-a-Judgecited by 1,423
  9. 2018Sanity Checks for Saliency Mapscited by 1,156
  10. 2019OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledgecited by 1,121
  11. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 599
  12. 2024LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methodscited by 465
5 more
  1. 2019Counterfactual Visual Explanationscited by 313
  2. 2025OmniGen2: Towards Instruction-Aligned Multimodal Generationcited by 253
  3. 2021Counterfactual Explanations Can Be Manipulatedcited by 113
  4. 2018Faithful Multimodal Explanation for Visual Question Answeringcited by 64
  5. 2024CounterCurate: Enhancing Physical and Semantic Visio-Linguistic Compositional Reasoning via Counterfactual Examplescited by 35

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 12.

12 papers.

  1. 2026On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMscited by 11
  2. 2026When to Call an Apple Red: Humans Follow Introspective Rules, VLMs Don'tcited by 1
  3. 2026Position, Not Provenance: Separating Reasoning Mediation from Sycophancy in Medical Vision-Language Modelscited by 1
  4. 2026BEVLM: Distilling Semantic Knowledge from LLMs into Bird's-Eye View Representationscited by 0
  5. 2026Revealing the Gap in Human and VLM Scene Perception through Counterfactual Semantic Saliencycited by 0
  6. 2026Faithful Action-unit Causal Reasoning for Counterfactually Faithful Emotion Explanationscited by 0
  7. 2026When Do Frozen VLMs Respond to Image-Free Object-Token Edits? An Answer-Key-Free Protocol and What It Revealscited by 0
  8. 2026Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection Guardrailscited by 0
  9. 2026Counterfactual Tests for Measuring Chain-of-Thought Faithfulness in Visual Language Modelscited by 0
  10. 2026Does a model's stated reason for rejecting a candidate do any work?cited by 0
  11. 2026AnswerMap: Faithful Spatial Interpretability of VLMs from Answer Posteriorscited by 0
  12. 2026Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026EDCT-Bench: Uncovering Faithfulness Gaps in VLMs via Explanation-Driven Counterfactual Testing
  2. 2024The Probabilities Also Matter: A More Faithful Metric for Faithfulness of Free-Text Explanations in Large Language Models
  3. 2023Benchmarking Faithfulness: Towards Accurate Natural Language Explanations in Vision-Language Tasks
  4. 2026Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning
  5. 2025CF-VLM:CounterFactual Vision-Language Fine-tuning
  6. 2025Believing without Seeing: Quality Scores for Contextualizing Vision-Language Model Explanations
  7. 2026CDH-Bench: A Commonsense-Driven Hallucination Benchmark for Evaluating Visual Fidelity in Vision-Language Models
  8. 2023What If the TV Was Off? Examining Counterfactual Reasoning Abilities of Multi-modal Language Models