# PhysEditBench: A Protocol-Conditioned Benchmark for Dense Physical-Map Prediction with Image Editors

arXiv: https://arxiv.org/abs/2605.13493 · HTML: https://mlsanity.com/paper/2605.13493/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 30 references that are in the map.
- First posted on arXiv 13 May 2026.

## Builds on (most cited first, 25 of 25)

- [InstructPix2Pix: Learning to Follow Image Editing Instructions](https://mlsanity.com/paper/2211.09800/) (2022)
- [Depth Anything 3: Recovering the Visual Space from Any Views](https://mlsanity.com/paper/2511.10647/) (2025)
- [MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing](https://mlsanity.com/paper/2306.10012/) (2023)
- [Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding](https://mlsanity.com/paper/2011.02523/) (2020)
- [Emu Edit: Precise Image Editing via Recognition and Generation Tasks](https://mlsanity.com/paper/2311.10089/) (2023)
- [Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction](https://mlsanity.com/paper/2409.18124/) (2024)
- [RGBleftrightarrowX: Image decomposition and synthesis using material- and lighting-aware diffusion models](https://mlsanity.com/paper/2405.00666/) (2024)
- [StableNormal: Reducing Diffusion Variance for Stable and Sharp Normal](https://mlsanity.com/paper/2406.16864/) (2024)
- [Infinite Photorealistic Worlds using Procedural Generation](https://mlsanity.com/paper/2306.09310/) (2023)
- [Infinigen Indoors: Photorealistic Indoor Scenes using Procedural Generation](https://mlsanity.com/paper/2406.11824/) (2024)
- [Rethinking Inductive Biases for Surface Normal Estimation](https://mlsanity.com/paper/2403.00712/) (2024)
- [Learning-based Inverse Rendering of Complex Indoor Scenes with Differentiable Monte Carlo Raytracing](https://mlsanity.com/paper/2211.03017/) (2022)
- [OpenRooms: An End-to-End Open Framework for Photorealistic Indoor Scene Datasets](https://mlsanity.com/paper/2007.12868/) (2020)
- [Intrinsic Image Diffusion for Indoor Single-view Material Estimation](https://mlsanity.com/paper/2312.12274/) (2023)
- [IntrinsicAnything: Learning Diffusion Priors for Inverse Rendering Under Unknown Illumination](https://mlsanity.com/paper/2404.11593/) (2024)
- [Colorful Diffuse Intrinsic Image Decomposition in the Wild](https://mlsanity.com/paper/2409.13690/) (2024)
- [Image Generators are Generalist Vision Learners](https://mlsanity.com/paper/2604.20329/) (2026)
- [IRIS: Inverse Rendering of Indoor Scenes from Low Dynamic Range Images](https://mlsanity.com/paper/2401.12977/) (2024)
- [IR3D-Bench: Evaluating Vision-Language Model Scene Understanding as Agentic Inverse Rendering](https://mlsanity.com/paper/2506.23329/) (2025)
- [MAIR: Multi-view Attention Inverse Rendering with 3D Spatially-Varying Lighting Estimation](https://mlsanity.com/paper/2303.12368/) (2023)
- [DNF-Intrinsic: Deterministic Noise-Free Diffusion for Indoor Inverse Rendering](https://mlsanity.com/paper/2507.03924/) (2025)
- [Boosting 3D Object Generation through PBR Materials](https://mlsanity.com/paper/2411.16080/) (2024)
- [Materialist: Physically Based Editing Using Single-Image Inverse Rendering](https://mlsanity.com/paper/2501.03717/) (2025)
- [MVInverse: Feed-forward Multi-view Inverse Rendering in Seconds](https://mlsanity.com/paper/2512.21003/) (2025)
- [ePBR: Extended PBR Materials in Image Synthesis](https://mlsanity.com/paper/2504.17062/) (2025)

## Abstract (arXiv)

Can general-purpose image editors predict physical maps from a single RGB image? General-purpose image editors differ from standard task-specific dense-prediction models: they do not directly take an image and output a physical map. Instead, they must be guided by prompts, examples, or image-based textual cues. To this end, we introduce PhysEditBench, a novel protocol-conditioned benchmark to evaluate and standardize image editors in dense physical-map prediction that covers five targets: depth, normal, albedo, roughness, and metallic maps. For evaluation data, we build a target-dependent benchmark substrate. We use OpenRooms-FF for depth, surface normal, albedo, and roughness, InteriorVerse as an additional source for depth, normal, albedo, and a new procedurally generated source for metallic maps. We curate the data with quality checks, valid-region masks, scene-level sampling, and lighting-based stress subsets to ensure reliable and diverse evaluation. For each target, PhysEditBench defines a fixed protocol that specifies the allowed input, expected output format, and scoring procedure. Each score, therefore, reflects the performance of a model under a specified protocol, rather than its best possible performance under all prompts or interaction modes. Experimental results show that specialized models remain much stronger on depth, normal, and albedo, and stronger image editors can produce more reasonable map-like outputs. For roughness and metallic, image editors can match or outperform specialized baselines on some scalar metrics, but they still suffer from structural errors, sparsity effects, and sensitivity to lighting.