metadata
license: apache-2.0
task_categories:
- image-to-3d
G4Splat: Geometry-Guided Gaussian Splatting with Generative Prior
Project page | Paper | Code
This repository contains the preprocessed data for G4Splat, a method presented at ICLR 2026 that integrates accurate geometry guidance with generative priors to enhance 3D scene reconstruction. G4Splat improves both geometric fidelity and appearance quality in observed and unobserved regions.
Dataset Description
The dataset includes preprocessed scans for training and evaluating G4Splat on several benchmarks:
- Replica
- ScanNet++
- DeepBlending
- DenseView
Usage
Data Structure
To use this data with the official G4Splat implementation, please download and unzip the files into a data folder following this structure:
└── G4Splat
└── data
├── replica
├── scan ...
├── scannetpp
├── scan ...
├── deepblending
├── scan ...
├── denseview
├── scan1
Training Command
Once the data is organized, you can run training and evaluation using the following command:
# Tested on A100 80GB GPU. You can add "--use_downsample_gaussians" to run on a 3090 24GB GPU.
python train.py -s data/DATASET_NAME/SCAN_ID -o output/DATASET_NAME/SCAN_ID --sfm_config posed --use_view_config --config_view_num 5 --select_inpaint_num 10 --tetra_downsample_ratio 0.25
Citation
@inproceedings{ni2026g4splat,
title={G4Splat: Geometry-Guided Gaussian Splatting with Generative Prior},
author={Ni, Junfeng and Chen, Yixin and Yang, Zhifei and Liu, Yu and Lu, Ruijie and Zhu, Song-Chun and Huang, Siyuan},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026}
}