LISU: Low-light Indoor Scene Understanding

Authors

Ning Zhang, Francesco Nex, Norman Kerle, George Vosselman

Introduction

This repository contains the datasets and source code of our framework LISU.

LISU is a cascade framework that joinly learns semantic segmentation and reflectance restoration on low-light indoor scenes.

Dataset: LLRGBD

LLRGBD is a newly collected dataset for low-light indoor scene understanding. It consists of one large-scale synthetic data set LLRGBD-synthetic and one small-scale real data set called LLRGBD-real. We provide pairs of low-light/normal-light images of each scene.

Usage:

Download Datasets:

LLRGBD-real: https://surfdrive.surf.nl/files/index.php/s/EM9sa5DgofvTHdi
LLRGBD-real-depth: https://surfdrive.surf.nl/files/index.php/s/xmSGFhn8gleXcaX
LLRGBD-synthetic: https://surfdrive.surf.nl/files/index.php/s/2GvpEMqQ2FVtGpV

Download checkpoint tarined on LLRGBD-real

https://surfdrive.surf.nl/files/index.php/s/vaWpfQ49nBkauWv

Evaluation on LLRGBD-real

python main.py --mode evaluation --data_path /path/to/your/dataset/ --pretrained-model-path /path/to/LISU_LLRGBD_real_best.pth.tar

Train on LLRGBD-real

python main.py --mode evaluation --data_path /path/to/your/dataset/ --pretrained_model_path None

Other usages

Please check the code for other usages.

Citation

If you use our code or datasets please cite:

@article{zhang2022lisu,
title={LISU: Low-light indoor scene understanding with joint learning of reflectance restoration},
author={Zhang, Ning and Nex, FC and Kerle, N and Vosselman, G},
journal={ISPRS journal of photogrammetry and remote sensing},
volume={183},
pages={470--481},
year={2022},
publisher={Elsevier}
}


@article{zhang2021towards,
title={Towards Learning Low-Light Indoor Semantic Segmentation with Illumination-Invariant Features},
author={Zhang, N and Nex, F and Kerle, N and Vosselman, G},
journal={The International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences},
volume={43},
pages={427--432},
year={2021},
publisher={Copernicus GmbH}
}