UniPhysGen-1.7B-Object
UniPhysGen-1.7B-Object is the object-level intrinsic physical grounding checkpoint from UniPhysGen. Given a full-object point cloud, it predicts object identity, category, dimensions, and mass. It does not require a target-part point cloud.
Model details
| Item | Value |
|---|---|
| Internal task name | object_level |
| Required geometry | Object point cloud only |
| Training lineage | UniPhysGen-1.7B-Init → UniPhysGen-1.7B-Physics → object-level fine-tuning |
| Training data | breezexian/UniPhys-40K |
| Evaluation data | UniPhys-Bench · UniPhys-Bench Part 2 |
| Tested Transformers version | 4.51.0 |
| Source code | breezexian/UniPhysGen |
| Paper | arXiv:2607.13586 |
Inputs and outputs
The recommended point-cloud format is .npz with aligned arrays:
point float32 [N, 3]
color uint8 [N, 3]
normal float32 [N, 3]
The structured prediction is:
{
"object_name": "cabinet",
"category": "Furniture/Cabinet",
"volume": [80.0, 40.0, 120.0],
"mass": 35.0
}
volume stores [length, width, height] in centimeters. mass is expressed
in kilograms.
Installation
The model has been tested on Linux with Python 3.11, PyTorch 2.4.1, CUDA 12.4,
and transformers==4.51.0.
[!IMPORTANT] Use
transformers==4.51.0. This is the tested version and is pinned by the UniPhysGen project metadata.
git clone https://github.com/breezexian/UniPhysGen.git
cd UniPhysGen/uniphysgen
conda create -n uniphysgen python=3.11 -y
conda activate uniphysgen
conda install -y -c nvidia/label/cuda-12.4.0 cuda-toolkit
python -m pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu124
python -m pip install -e ".[train]"
bash scripts/install_cuda_extensions.sh
For inference only, replace python -m pip install -e ".[train]" with
python -m pip install -e ..
The model uses a custom Transformers architecture. Install UniPhysGen before
loading the checkpoint; a generic transformers.pipeline is not supported.
Inference
CUDA_VISIBLE_DEVICES=0 python inference_batch_intrinsic_physics_object.py \
--model_path breezexian/UniPhysGen-1.7B-Object \
--object_pcd examples/object.npz \
--output outputs/intrinsic_physics_object.json
For batch inference, pass a JSON list with --input_json. No --part_pcd
argument is needed for this task.
Evaluation
python -m eval intrinsic_physics_object PREDICTIONS \
--output intrinsic_physics_object_metrics.json
See Table 2 of the paper for the UniPhys-Bench results and the main project README for the complete evaluation protocol.
Intended use and limitations
This checkpoint is intended for research on object-level physical grounding and for proposing approximate object metadata for downstream simulation. Predicted dimensions and mass are estimates, not calibrated measurements.
Performance may degrade for incomplete scans, incorrect scale, sparse or noisy point clouds, unusual materials, composite objects, or categories outside the training distribution. Validate predictions before using them in robotics, safety-critical systems, or engineering workflows.
License
The model weights are released under the Creative Commons Attribution-NonCommercial 4.0 International license. Commercial use is not permitted under this license. The UniPhysGen source code is licensed separately under Apache-2.0. See the included license for Qwen3 and Sonata attribution.
Citation
@article{li2026uniphysgen,
title = {UniPhysGen: Unified Physical Grounding for Simulation-Ready 3D Assets},
author = {Li, Xian and Wei, Rong and Yang, Lujie and Huang, Haolin and Fang, Junyuan and Tang, Siliang and Xiao, Jun and Tang, Rui and Li, Juncheng},
journal = {arXiv preprint arXiv:2607.13586},
year = {2026}
}