SHARE: Pose-Free 3D Gaussian Splatting
via Shape-Ray Estimation

Youngju Na*  ·  Taeyeon Kim*  ·  Jumin Lee  ·  Kyu Beom Han
Woo Jae Kim  ·  Sung-Eui Yoon

ICIP 2025 Best Student Paper Award

Paper | Project Page

📰 News

• To Do: Release SHAREv2 (SHARE++), which adopts 3D foundation models in our pipeline.
• (10/21) Release checkpoint links
• (10/21) Initial code release
• (10/15) We received the ICIP 2025 Best Student Paper Award!

🧩 Installation

To get started, create a conda virtual environment using Python 3.10+ and install the requirements:

conda create -n share python=3.10
conda activate share
pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

📦 Datasets

SHARE is trained and evaluated on DTU and RealEstate10K datasets.

• DTU dataset follows the preprocessing and download procedure of VolRecon.
• RealEstate10K dataset follows the setup of MVSplat.

Please refer to the respective repositories for detailed dataset preparation instructions.

• To Do: add DL3DV dataloader for training and evaluation.

📥 Checkpoints

Pretrained SHARE model checkpoints are available via google form or email (to first author).

The folder contains checkpoints for datasets trained on DTU and RealEstate10K, respectively.

To use a checkpoint, download the relevant file and specify its path in your configuration or command line arguments as described in the training and evaluation scripts.

---

🚀 Running the Code

Training and Evaluation

To run SHARE on each dataset, please refer to the command templates provided in the following files:

• commands_dtu.md – training and evaluation on DTU / BlendedMVS
• commands_re10k.md – training and evaluation on RealEstate10K / ACID
• commands_cross_dataset.md – for cross-dataset evaluation

All scripts are located in the project root and designed for easy reproducibility of the results reported in the paper.

---

📈 Cross-Dataset Evaluation

We provide ready-to-run scripts for evaluating cross-domain generalization using commands_cross_dataset.md.
Please refer to the file for detailed instructions and dataset pair configurations.

🧪 Ablation Study

We provide ready-to-run scripts for ablation studies in commands_ablation.md.
Please refer to the file for detailed instructions and experiment configurations.

---

🧠 BibTeX

@article{na2025share,
  title={SHARE: Pose-Free 3D Gaussian Splatting via Shape-Ray Estimation},
  author={Na, Youngju and Kim, Taeyeon and Lee, Jumin and Han, Kyu Beom and Kim, Woo Jae and Yoon, Sung-eui},
  journal={arXiv preprint arXiv:2505.22978},
  year={2025}
}

---

🙏 Acknowledgements

This project builds upon the foundations of
MVSplat (ECCV 2024) and UFORecon (CVPR 2024),
and incorporates components adapted from these excellent works.
We sincerely thank the original authors for their open-source contributions.

---