metadata
license: apache-2.0
tags:
- radiance-fields
- novel-view-synthesis
- 3d-reconstruction
- triangle-splatting
- eccv-2026
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},
}