Memory Efficient Voxelized Renderable Neural 3D Spatial Representation for Vision-Based Robotics

Howoong Jun, Seongbo Ha, Jaewon Lee, Hyeonwoo Yu, and Songhwai Oh

IEEE Robotics and Automation Letters, 2026

Paper | Video

Environments

We provide docker repository for this work.

docker pull howoongjun/3dgs:3dsr

Datasets

• Replica

  • Download

  bash download_replica.sh

  • Folder structure

  Replica
      - office0
          - results (contain rgbd images)
              - frame000000.jpg
              - depth000000.jpg
              ...
          traj.txt
      ...

• TUM

  • Download

  bash download_tum.sh

• HM3D

  • The sample data for HM3D will be uploaded soon

Prerequisites

Before running the code, please download the checkpoint for the upsampling network and save it in the super_resolution/checkpoints folder. The link provides checkpoints for the Replica and TUM datasets.

• Folder structure

3dsr
    - super_resolution
        - checkpoints
            - checkpoint_srresnet_voxel_x4_office0_300x170.pth.tar
            - checkpoint_srresnet_voxel_x4_office1_300x170.pth.tar
            ...

Run

You can try rendering with sample data using demo notebook. Additionally, you can review the evaluation results on image quality metrics, including PSNR, LPIPS, and SSIM.

• [Demo] Rendering.ipynb

BibTex

@InProceedings{jun20253dsr,
    author  = {Jun, Howoong and Ha, Seongbo and Lee, Jaewon and Yu, Hyeonwoo and Oh, Songhwai},
    title   = {Memory Efficient Voxelized Renderable Neural 3D Spatial Representation for Vision-Based Robotics},
    journal = {IEEE Robotics and Automation Letters},
    year    = {2026}
}