README.md

metadata

license: mit
task_categories:
  - image-to-3d

EKS Dataset

This repository contains the data and assets used in the paper Affine-Equivariant Kernel Space Encoding for NeRF Editing, which introduces a hybrid model for interactive editing of Neural Radiance Fields.

Project page | Code

Dataset content

In the following folders you can find:

• blender - our animations with scripts used to generate them
• configs - configuration files used during the experimetns shown in our paper
• data - contains NeRF Synthetic + Our Assets wit appriopriate sparse_pc.ply used for initialization of the system
• permuto_SDF_models - meshes generated with PermutoSDF that we have used for driving the Gaussians

Dataset lacks the data for Mip-NeRF 360 dataset which can be downloaded from here and also fox which can be found here.
Additionally Mip-NeRF 360 shoud be processed with:

# Do this for every dataset in the folder
cd <dataset_folder>
ns-process-data images --data . --output-dir . --skip-colmap --skip-image-processin --colmap-model-path sparse/0

Bugs in nerfstudio 1.1.4

There were a few bugs in nerfstudio we needed to fix in order to train on Mip-NeRF 360 dataset:

File: nerfstudio/exporter/exporter_utils.py

# Lines 166-172

# Change from:
if crop_obb is not None:
    mask = crop_obb.within(point)
point = point[mask]
rgb = rgb[mask]
view_direction = view_direction[mask]
if normal is not None:
    normal = normal[mask]

# To:
if crop_obb is not None:
    mask = crop_obb.within(point)
    point = point[mask]
    rgb = rgb[mask]
    view_direction = view_direction[mask]
    if normal is not None:
        normal = normal[mask]

File: nerfstudio/model_components/ray_generators.py

# Lines 49-50

# Change from:
y = ray_indices[:, 1]  # row indices
x = ray_indices[:, 2]  # col indices

# To:
y = torch.clamp(ray_indices[:, 1], 0, self.image_coords.shape[0] - 1)  # row indices
x = torch.clamp(ray_indices[:, 2], 0, self.image_coords.shape[1] - 1)  # col indices

File: nerfstudio/utils/eval_utils.py

# Line 62

# Change from:
loaded_state = torch.load(load_path, map_location="cpu")

# To:
loaded_state = torch.load(load_path, map_location="cpu", weights_only=False)

📄 Citation

If you use our data, please cite:

@misc{zielinski2025eks,
  title     = {Affine-Equivariant Kernel Space Encoding for NeRF Editing},
  author    = {Miko\l{}aj Zieli\'{n}ski and Krzysztof Byrski and Tomasz Szczepanik and Dominik Belter and Przemys\l{}aw Spurek},
  year      = {2025},
  eprint    = {2508.02831},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url       = {https://arxiv.org/abs/2508.02831}
}