PyTorch implementation of Neural Poisson Solver.
We provide a conda environment setup file including all of the above dependencies. Create the conda environment Neural Poisson Solver by running:
conda create -n neural-poisson-solver python=3.8
conda activate neural-poisson-solver
pip install -r requirements.txt
For 2D scene blending tasks, we employ DINER as the backbone network.
You need to prepare the following data and place them in the data/2d/
folder. The directory structure should look like this:
data/2d/
├── scene_1/
│ ├── src.pth # Source scene's INR model
│ ├── tgt.pth # Target scene's INR model
│ ├── roi.png # Blending region
│ ├── cfg.npy # Blending center coordinates
└── ...
export PYTHONPATH=$(pwd)
CUDA_VISIBLE_DEVICES=0 python src/blending/blend_2d.py --save_dir results/2d/scene_1/ --root_dir data/2d/scene_1/ --use_numpy False
For 3D scene blending tasks, we employ NeRF as the backbone network.
You need to prepare the following data and place them in the data/3d/
folder. The directory structure should look like this:
data/3d/
├── scene_1/
│ ├── src.tar # Source scene's NeRF model
│ ├── tgt.tar # Target scene's NeRF model
│ ├── roi.pt # Blending region
│ ├── cam.json # Camera parameters
└── ...
export PYTHONPATH=$(pwd)
sh src/models/nerf/install_ext.sh
CUDA_VISIBLE_DEVICES=0 python src/blending/blend_3d.py --save_dir results/3d/scene_1/ --root_dir data/3d/scene_1/
@InProceedings{10.1007/978-3-031-72989-8_15,
author="Wu, Delong
and Zhu, Hao
and Zhang, Qi
and Li, You
and Ma, Zhan
and Cao, Xun",
editor="Leonardis, Ale{\v{s}}
and Ricci, Elisa
and Roth, Stefan
and Russakovsky, Olga
and Sattler, Torsten
and Varol, G{\"u}l",
title="Neural Poisson Solver: A Universal and Continuous Framework for Natural Signal Blending",
booktitle="Computer Vision -- ECCV 2024",
year="2025",
publisher="Springer Nature Switzerland",
address="Cham",
pages="259--275",
}