detris-models

DETRIS — Deformable Triangle Splatting (pretrained models)

Pretrained models for Deformable Triangle Splatting: Flexible Primitives for Real-Time Radiance Field Rendering (ECCV 2026).

Oriol Jiménez-Ayguadé · Antonio Agudo — Institut de Robòtica i Informàtica Industrial, CSIC-UPC, Barcelona.

Project page · arXiv · Code

Video

Contents

Final (30 000-iteration) checkpoints for 11 scenes:

• Mip-NeRF 360 (9): bicycle, bonsai, counter, flowers, garden, kitchen, room, stump, treehill
• Tanks & Temples (2): train, truck

Each scene folder contains:

<scene>/
├── cfg_args                         # training configuration
├── cameras.json                     # camera parameters
├── input.ply                        # initial point cloud
├── results.json, per_view.json      # evaluation metrics
├── events.out.tfevents.*            # training log (TensorBoard)
└── point_cloud/iteration_30000/
    ├── point_cloud_state_dict.pt    # trained DETRIS primitives
    └── hyperparameters.pt

Usage

# 1. clone the code
git clone --recursive https://github.com/orioljim1/detris
cd detris

# 2. download & extract this release
#    (e.g. with the huggingface_hub CLI)
hf download Orioljim/detris-models detris_models.zip --local-dir .
unzip detris_models.zip

# 3. render a scene (see the code repo README for full options)
python render.py -m main_results/mipnerf360/garden

See the code repository for training, evaluation, and rendering instructions.

License

Released under the Apache License 2.0. See LICENSE.md in the code repository.

Citation

@inproceedings{JimenezAyguade2026Deformable,
  title     = {Deformable Triangle Splatting: Flexible Primitives for Real-Time Radiance Field Rendering},
  author    = {Jiménez-Ayguadé, Oriol and Agudo, Antonio},
  booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
  year      = {2026},
}