EAG-PT: Emission-Aware Gaussians and Path Tracing for Diffuse Indoor Scene Reconstruction and Editing

Project Page | Paper | EAG-PT-tracer

Xijie Yang, Mulin Yu*, Changjian Jiang, Kerui Ren, Tao Lu, Jiangmiao Pang, Dahua Lin, Bo Dai*, Linning Xu

Get Started

Download Code

git clone https://github.com/InternRobotics/EAG-PT.git
git clone https://github.com/InternRobotics/EAG-PT-tracer.git

Prepare Data

Please check _data/README.md to download multi-view indoor scenes used in the paper.

Setup Environment

We tested the code on Ubuntu 22.04 with NVIDIA RTX 4090, Driver 570.153.02, CUDA 12.8, nvcc 12.8, and OptiX 7.7. For machine specification, please check _docs/notes.md. The code should also work on other NVIDIA RTX cards with higher CUDA versions.

# (optionally) `libgl-dev` fixes "OSError: libGL.so.1: cannot open shared object file: No such file or directory". `libnvidia-gl-570` adds `/lib/x86_64-linux-gnu/libnvoptix.so.1` and `/usr/share/nvidia/nvoptix.bin`, fixing "RuntimeError: Could not initialize OptiX!".
apt install libgl-dev libnvidia-gl-570

conda create -y -n EAG-PT python=3.11
conda activate EAG-PT

pip install torch==2.9.0 torchvision==0.24.0 --index-url https://download.pytorch.org/whl/cu128
pip install numpy==2.2.6 einops==0.8.1 tqdm==4.67.1 opencv-python-headless==4.12.0.88 OpenImageIO==3.0.4.0 mitsuba==3.7.1 plyfile==1.1.3 open3d==0.19.0 lpips==0.1.4 flip-evaluator==1.7

Install Tracer

cd EAG-PT-tracer/

pip install ninja==1.13.0
rm -rf *_optix/ptx/ build/ *.egg-info/; TORCH_CUDA_ARCH_LIST=8.9 pip install . -v --no-build-isolation

• arch 8.9 is for RTX 4090, for other gpus, check https://developer.nvidia.com/cuda-gpus. arch should also be changed in setup.py command_nvcc (and in eag_pt_tracer_optix/program/generate-ptx.sh).
• If the tracer cannot be installed and used successfully, please open an issue in EAG-PT-tracer.

Run Code

cd EAG-PT/

# Reconstruct the scene.
_scripts/0-and-1-reconstruction.sh
# Edit and render the scene.
_scripts/editing-and-rendering.sh
# Bake light on edited scene.
_scripts/light-baking.sh

• Scenes, scenarios, parameters could be manually changed in the script.
• Scene editing may require some assets from other reconstructed scenes.
• Results will be saved in _output/.
• For viewing EXR files, use the VS Code extension VERIV or local software tev.

Work in Progress

• Add tracer usage examples.
• Release math details in the tracer.
• Tidy up data capture process.

Feedback and Contributions

Issues, pull requests, and suggestions are welcome. If you find EAG-PT useful, please consider starring this repository and citing our paper 😉.

Citation

@misc{EAG-PT-2026-XijieYang,
    title={EAG-PT: Emission-Aware Gaussians and Path Tracing for Diffuse Indoor Scene Reconstruction and Editing},
    author={Xijie Yang and Mulin Yu and Changjian Jiang and Kerui Ren and Tao Lu and Jiangmiao Pang and Dahua Lin and Bo Dai and Linning Xu},
    year={2026},
    eprint={2601.23065},
    archivePrefix={arXiv},
    primaryClass={cs.GR},
    doi={https://doi.org/10.1145/3799902.3811054},
    url={https://arxiv.org/abs/2601.23065},
}