In this paper, we propose a new method, named Sparis, for indoor surface reconstruction from sparse views. Specifically, we investigate the impact of monocular priors on sparse scene reconstruction, introducing a novel prior based on inter-image matching information. Our prior offers more accurate depth information while ensuring cross-view matching consistency.

conda create -n sparis python=3.8
conda activate sparis
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install -r requirements.txt

You can download the sparse-view ScanNet and Replica dataset from this link and put them in the ./data directory. The data structure should be like:

|-- code
|-- data
    |-- ScanNet
        |-- <scan_name, e.g., scan710>
            |-- cameras.npz
            |-- 000000_depth.npy
            |-- 000000_normal.npy
            |-- 000000_rgb.png
            ...
            |-- match_roma
                |-- match_dict_0.npz
                ...
    |-- Replica
    ...

cd code

# ScanNet
CUDA_VISIBLE_DEVICES=0 python training/exp_runner.py --conf confs/scannet_mlp.conf --scan_id <scan_id, e.g., 710>

# Replica
CUDA_VISIBLE_DEVICES=0 python training/exp_runner.py --conf confs/replica_mlp.conf --scan_id <scan_id, e.g., 1>

If you find our work useful in your research, please consider citing:

@inproceedings{wu2025sparis,
    title={Sparis: Neural Implicit Surface Reconstruction of Indoor Scenes from Sparse Views},
    author={Yulun Wu and Han Huang and Wenyuan Zhang and Chao Deng and Ge Gao and Ming Gu and Yu-Shen Liu},
    booktitle={AAAI Conference on Artificial Intelligence},
    year={2025}
}

This implementation is built upon MonoSDF. Thanks to the authors for their great work.