WIPES: Wavelet-based Visual Primitives

ICCV 2025

Wenhao Zhang^1,*, Hao Zhu^1,*, Delong Wu^1,*, Di Kang², Linchao Bao², Xun Cao¹, Zhan Ma¹

¹Nanjing University, ²Tencent, ^*Equal contibution

Project Page | Paper

We propose WIPES, a universal Wavelet-based vIsual PrimitivES for representing multi-dimensional visual signals. Building on the spatial-frequency localization advantages of wavelets, WIPES effectively captures both the low-frequency "🌴🌴forest🌴🌴" and the high-frequency "trees.🌳"

This repository provides the code for several applications:

• 2D Image Fitting: Demonstrates the model's ability to represent 2D images. Our Image fitting experiments are built upon the GaussianImage codebase
• 5D NVS: Our Static NVS experiments are built upon the gaussian-splatting codebase.

Setup

wipes_image

conda create -n wipes_image python=3.8
conda activate wipes_image
cd wipes_image
pip install -r requirements.txt
cd wipesplat
pip install .[dev]

wipes_splatting

cd wipes_splatting
conda env create --file environment.yml
conda activate wipes_splatting

Training

Image Fitting

cd wipes_image
bash scripts/wipesimage_cholesky/toy_exp.sh.sh 

5D NVS

cd wipes_splatting
bash scripts/train_single.sh

Citation

@InProceedings{Zhang_2025_ICCV,
    author    = {Zhang, Wenhao and Zhu, Hao and Wu, Delong and Kang, Di and Bao, Linchao and Cao, Xun and Ma, Zhan},
    title     = {WIPES: Wavelet-based Visual Primitives},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {27338-27347}
}