README.md

---

license: cc-by-4.0
tags:
  - digital-twins
  - scene-graphs
  - healthcare
  - structured-scene-states
  - reproducibility
  - procedural
---


# MedPhyGraph — Procedural Support-Graph Data

**MedPhyGraph: Counterfactual Support-Graph Maintenance for Dynamic Built-Environment Digital Twins**

| Resource | Link |
|----------|------|
| **Project Website** | https://medphygraph.github.io/ |
| **Model (CF-SupportNet)** | https://huggingface.co/MedPhyGraph/CF-SupportNet |
| **Paper** | \<PAPER_URL\> |
| **Code** | \<GITHUB_URL\> |

## Dataset description

This repository contains structured procedural training and evaluation data used in
the MedPhyGraph paper. It supports the paper's method contribution and is not
presented as a standalone benchmark contribution.

All redistributed artifacts are **MedPhyGraph-owned procedural structured JSON**.
No NVIDIA Isaac Sim / Isaac for Healthcare assets, USD binaries, meshes, textures,
or Isaac-derived structured scene files are included.

This repository contains the **Procedural subset (136 of 217 cases)** of the
expanded transfer suite. The paper additionally reports **81 Isaac for Healthcare

cases**, which are **not redistributed here**.

### Paper evaluation vs. this public release

| Suite | Paper (reported) | This repository |
|-------|------------------|-----------------|
| Expanded support-transfer cases | **217** total | **136** procedural only |
| Procedural transfers | 136 | **136** |
| Isaac for Healthcare transfers | 81 | **0** (excluded) |

The public expanded-transfer subset is **`expanded_transfer_procedural_v1/`**.

It is **not** the complete 217-case paper suite.



## Components



### 1. `training_data/`



Paper-frozen CF-SupportNet training/evaluation feature dataset (`dataset_hard`).



| File | Description |

|------|-------------|

| `dataset.json` | 533 analytic counterfactual edge samples |

| `split.json` | Scene-level train/val/test split (324 / 101 / 108 samples) |



**Provenance:** Generated from procedurally authored `hard_000`–`hard_019` scenes via

`hard_scenes.py` and `generate_counterfactual_dataset.py`. No Isaac-derived scenes.



### 2. `procedural_scenes/`



Structured **Procedural** evaluation states (TwinWorld Phase-2 dynamic corpus).



| Property | Value |

|----------|-------|

| `scene.json` directories | 54 |

| Geometry | Procedural AABB proxies (`proxy://aabb/...`) |



### 3. `expanded_transfer_procedural_v1/`



Public procedural subset of the paper's expanded transfer evaluation suite.



| Path | Description |

|------|-------------|

| `predeclared_manifest.json` | Eligibility manifest (procedural cases only) |

| `targets.json` | Ground-truth targets (**136** procedural cases) |

| `scenes/procedural/` | Authored structured scene states for each case |



## NVIDIA Isaac for Healthcare: reproduction scope



This release distinguishes between the fully public Procedural evaluation and

the Isaac for Healthcare evaluation, which depends on separately licensed

NVIDIA software and assets.



### 1. Procedural experiments — fully reproducible



The procedural training data, frozen train/validation/test split, structured

procedural scenes, CF-SupportNet checkpoints, and the Procedural subset of the

expanded transfer evaluation are publicly released.



The public expanded-transfer subset contains 136 Procedural cases. These

artifacts support end-to-end reproduction of the released Procedural

experiments.



### 2. Isaac for Healthcare — NVIDIA software and assets are required



The MedPhyGraph paper also reports experiments in NVIDIA Isaac for Healthcare

environments.



NVIDIA Isaac for Healthcare assets, USD/USDA/USDC files, meshes, textures, and

Isaac-derived structured scene states are not redistributed in this release.



Users interested in running MedPhyGraph with Isaac for Healthcare must obtain

the required NVIDIA Isaac Sim / Isaac for Healthcare software and assets

directly from NVIDIA under the applicable NVIDIA license terms.



The public MedPhyGraph codebase contains relevant components for working with

locally installed Isaac environments, but the complete paper-frozen Isaac scene

generation workflow is not currently packaged as a standalone end-to-end public pipeline.



### 3. Scope of the paper-frozen Isaac evaluation



The paper's expanded transfer evaluation contains 217 cases in total:



- 136 Procedural cases

- 81 Isaac for Healthcare cases



This repository redistributes the 136 Procedural cases only. The 81 Isaac cases

and their corresponding structured scene states are not included.



Accordingly, the exact Isaac results reported in the paper should not be

expected to be reproducible from the currently released public artifacts alone.

Reconstructing that portion requires the appropriate NVIDIA software and assets

as well as additional MedPhyGraph generation tooling.



This limitation does not affect reproduction of the publicly released

Procedural experiments.



## Expected layout after download



Mirror paths relative to the MedPhyGraph code repository:



```

outputs/dyphygraph_health/dataset_hard/              # from training_data/

outputs/dyphygraph_health/scenes_twinworld_phase2/   # from procedural_scenes/

outputs/reviewer_response/expanded_transfer_procedural_v1/

```



## Integrity



- `SHA256SUMS.txt` — per-file SHA256 digests

- `MANIFEST.json` — public metadata index (relative paths only)



## Not included



| Artifact | Reason |

|----------|--------|

| `scenes_twinworld_i4h_dynamic/` | Isaac-derived structured states — not redistributed |

| Isaac expanded-transfer cases (81) | Isaac-derived — excluded from public subset |

| NVIDIA / i4h USD assets | Third-party license |

| Canonical rebuild `outputs/medphygraph/` | Different research track |



## Citation



```bibtex

@inproceedings{gholizadeh2026medphygraph,

  title={MedPhyGraph: Counterfactual Support-Graph Maintenance for Dynamic Built-Environment Digital Twins},

  author={Gholizadeh HamlAbadi, Kamran and Vahdati, Monica and El Saddik, Abdulmotaleb},

  booktitle={ECCV 2026 TwinWorld Workshop},

  year={2026}

}

```



## License



This dataset is released under the **Creative Commons Attribution 4.0 International

License (CC BY 4.0)**. See the `LICENSE` file in the MedPhyGraph code repository

or https://creativecommons.org/licenses/by/4.0/.



Third-party NVIDIA Isaac for Healthcare assets are **not** included and are **not**

covered by this license.