# Do Vision Models Learn Physical Constraints or Rendering Shortcuts? A Counterfactual Benchmark for Grounded Physical Consistency

arXiv: https://arxiv.org/abs/2610.09205 · HTML: https://mlsanity.com/paper/2610.09205/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 24 references that are in the map.
- First posted on arXiv 6 Oct 2026.

## Builds on (most cited first, 10 of 10)

- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency](https://mlsanity.com/paper/2508.18265/) (2025)
- [FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models](https://mlsanity.com/paper/2410.02761/) (2024)
- [NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples](https://mlsanity.com/paper/2410.14669/) (2024)
- [ForgeryGPT: A Multimodal LLM for Interpretable Image Forgery Detection and Localization](https://mlsanity.com/paper/2410.10238/) (2024)
- [ForenX: Towards Explainable AI-Generated Image Detection with Multimodal Large Language Models](https://mlsanity.com/paper/2508.01402/) (2025)
- [PhyDetEx: Detecting and Explaining the Physical Plausibility of T2V Models](https://mlsanity.com/paper/2512.01843/) (2025)
- [Is This Edit Correct? A Multi-Dimensional Benchmark for Reasoning-Aware Image Editing](https://mlsanity.com/paper/2606.05172/) (2026)
- [PhyEditBench: A Real-World Multi-Stage Benchmark for Physics-Aware Image Editing](https://mlsanity.com/paper/2606.26551/) (2026)
- [SPR-128K: A New Benchmark for Spatial Plausibility Reasoning with Multimodal Large Language Models](https://mlsanity.com/paper/2505.23265/) (2025)

## Abstract (arXiv)

Modern image editing models can satisfy a text instruction while breaking the physics of the edited scene. A new object may cast no shadow, a mirror may fail to reflect visible geometry, or an object may float above a surface that should support it. We study physical plausibility diagnosis, detecting whether an edited image violates scene physics, naming the violation type, localizing the affected region, and explaining the failure in language. We introduce a counterfactual benchmark whose controlled synthetic component uses Mitsuba~3 to generate 5,500 images from 500 scene families. Each family contains one clean image and ten matched violations involving shadows, reflection, support, surface response, and occlusion. The renderer pipeline provides category labels, affected-region masks and boxes, scene metadata, and explanation targets. We use LLaVA-1.5-7B, Qwen2.5-VL-7B, and InternVL3.5-8B as diagnostic baselines rather than proposed methods. On a 1,650-image synthetic test set, the adapted baselines reach 64.0--67.8% category macro-F1 on standard held-out scenes. For LLaVA-1.5-7B, category macro-F1 falls from 64.0% on the standard split to 40.8% under intervention shift. This gap shows that high in-distribution accuracy partly reflects cues tied to rendering and counterfactual construction.