From Multi-View Images to Text-Guided Neural Surface Edits
Official Implementation

NeuSEditor enables text-guided neural surface editing directly from multi-view images.
This repository contains the official code and instructions to reproduce our results.

NeuSEditor is tested on NVIDIA RTX A6000, A100, and H100 GPUs.

• Recommended: CUDA 11.8 or 12.1, PyTorch 2.1 (see environment.yml for details).

To get started, create and activate the Conda environment:

conda env create -f environment.yml
conda activate neus_editor

You can verify that your GPU is accessible and CUDA versions are correct with:

python -c "import torch; print(torch.cuda.is_available(), torch.version.cuda, torch.__version__)"

NeRF-Synthetic (Blender)

• Download: Google Drive Link
• Usage: Place the downloaded folders under load/, e.g.:

  load/nerf_synthetic/lego
  

DTU (Preprocessed by NeuS)

• Download: Google Drive Link
• Usage: Use these scans with the DTU config.

Below are example commands to launch training for several supported datasets. Adjust command-line flags as needed.

python launch.py --config configs/blender.yaml --gpu 0 --train tag=example

python launch.py --config configs/dtu.yaml --gpu 0 --train

python launch.py --config configs/dtu.yaml --gpu 0 \
  tag=scan24_church_gs350 \
  diffusion.guidance_scale=350 \
  diffusion.tgt_prompt="make it a church" \
  dataset.root_dir=../data/DTU/scan24 \
  --train

python launch.py --config configs/colmap.yaml --gpu 0 --train

To get COLMAP data from custom images, you should have COLMAP installed (see here for installation instructions). Then put your images in the images/ folder, and run scripts/imgs2poses.py specifying the path containing the images/ folder. For example:

python scripts/imgs2poses.py ./load/images # images are in ./load/images

This part is adopted from Instant-NSR-PL; please refer to that repository for detailed usage and best practices: https://github.com/bennyguo/instant-nsr-pl

Parts of this codebase and scripts are inspired by or adapted from:

• Instant Neural Surface Reconstruction (Instant-NSR/Instant-NGP Lightning): https://github.com/bennyguo/instant-nsr-pl
• PDS: https://github.com/KAIST-Visual-AI-Group/PDS
• threestudio: https://github.com/threestudio-project/threestudio

If you find this work useful, please cite:

@inproceedings{ibrahimli2026neuseditor,
  title = {NeuSEditor: From Multi-View Images to Text-Guided Neural Surface Edits},
  author = {Ibrahimli, Nail and Kooij, Julian and Nan, Liangliang},
  booktitle = {International Conference on 3D Vision (3DV)},
  year = {2026}
}

This project is released under the terms of the license found in LICENSE.