Notre article a été accepté comme article de conférence à l'ECCV 2022 !
ISMVSNet, alias MVSNet basé sur l'échantillonnage d'importance, est une méthode de reconstruction multi-vues simple mais efficace.
Ce référentiel fournit une implémentation d'IS-MVSNet basée sur Mindspore. Vous pouvez suivre et regarder ce dépôt pour d'autres mises à jour.
# Centos 7.9.2009 is recommended.
# CUDA == 11.1, GCC == 7.3.0, Python == 3.7.9
conda create -n ismvsnet python=3.7.9
conda install mindspore-gpu=1.7.0 cudatoolkit=11.1 -c mindspore -c conda-forge # Install mindspore == 1.7.0
pip install numpy, opencv-python, tqdm, Pillow
conda activate ismvsnet
Les poids pré-entraînés pour le squelette sont déjà placés sous ./weights
. Les poids des étapes 1 à 3 peuvent être téléchargés à partir des poids pré-entraînés.
DATAROOT
└───data
| └───tankandtemples
| └───intermediate
| └───Playground
| │ └───rmvs_scan_cams
| │ │ 00000000_cam.txt
| │ │ 00000001_cam.txt
| │ │ ...
| │ └───images
| │ │ 00000000.jpg
| │ │ 00000001.jpg
| │ │ ...
| │ └───pair.txt
| │ └───Playground.log
| └───Family
| └───...
| └───advanced
└───weights
└───src
└───validate.py
└───point_cloud_generator.py
python validate.py
Les prédictions de profondeur seront enregistrées dans 'results/{dataset_name}/{split}/deepth'.
python point_cloud_generator.py
Les nuages de points fusionnés seront enregistrés dans 'results/{dataset_name}/{split}/points'
Si vous pensez que ce référentiel est utile, pensez à citer notre article :
@InProceedings{ismvsnet,
author="Wang, Likang
and Gong, Yue
and Ma, Xinjun
and Wang, Qirui
and Zhou, Kaixuan
and Chen, Lei",
editor="Avidan, Shai
and Brostow, Gabriel
and Ciss{'e}, Moustapha
and Farinella, Giovanni Maria
and Hassner, Tal",
title="IS-MVSNet:Importance Sampling-Based MVSNet",
booktitle="Computer Vision -- ECCV 2022",
year="2022",
publisher="Springer Nature Switzerland",
address="Cham",
pages="668--683",
abstract="This paper presents a novel coarse-to-fine multi-view stereo (MVS) algorithm called importance-sampling-based MVSNet (IS-MVSNet) to address a crucial problem of limited depth resolution adopted by current learning-based MVS methods. We proposed an importance-sampling module for sampling candidate depth, effectively achieving higher depth resolution and yielding better point-cloud results while introducing no additional cost. Furthermore, we proposed an unsupervised error distribution estimation method for adjusting the density variation of the importance-sampling module. Notably, the proposed sampling module does not require any additional training and works reasonably well with the pre-trained weights of the baseline model. Our proposed method leads to up to {$}{$}20{backslash}times {$}{$}20{texttimes}promotion on the most refined depth resolution, thus significantly benefiting most scenarios and excellently superior on fine details. As a result, IS-MVSNet outperforms all the published papers on TNT's intermediate benchmark with an F-score of 62.82{%}. Code is available at github.com/NoOneUST/IS-MVSNet.",
isbn="978-3-031-19824-3"
}