Indoor Localization Dataset V1.0 (Dept. & Metro St.)

NAVER LABS' open data set keeps being updated. We hope to grow together with researchers by sharing our latest data. On this page, you can check the composition and detailed information of the individual dataset 'Indoor Localization Dataset V1.0 (Dept. & Metro St.)'.

Dataset Description

NAVER LABS’ indoor localization dataset includes various characteristics of daily indoor spaces, aiming to support practical research in the field of vision technology. Locations in department stores (Dept.) and metro stations (Metro St.), with varied sizes (5,250~20,879m2) and unique characteristics, were selected, and more than 130,000 images and LiDAR data were collected from crowded environments. In addition, our dataset allows algorithm verification for changing environments, with data collected in intervals of maximum 4 months. The collected data will be provided in kapture format, with ground truth pose and image features generated through the mapping process.

Dept. dataset consists of data collected from 3 different floors. The dataset from 1F with cosmetics and jewelry stores and 4F with clothing stores contain bright lighting and much reflected information. On the other hand, the dataset from B1 where the food court is located contains dark images due to the relatively insufficient lighting condition.
As Metro St. dataset is collected from one of the most crowded stations in Seoul, it is suitable in evaluating algorithm robustness on images with occlusions due to multiple crowds. The B1 dataset is the largest dataset, including information on stores in the metro station and ticket gates. B2 dataset allows evaluation of an algorithm's performance in differentiating two similar metro platforms.

For easier evaluation of localization algorithms, NAVER LABS has collected data multiple times in each space. Each dataset consists of mapping data and query data. Query data is again categorized into test set and validation set. For detailed explanation, please refer to our paper.

Table 1. Dataset information

Figure 1. Point clouds of datset (white region: test set, red region: validation set)

Mapping Device

NAVER LABS used M1X to scan the department store and metro station. M1X is a wheel-based mobile device that moves in indoor spaces to collect data using 2 Velodyne LiDAR sensors, 6 industrial cameras, and 4 smartphone cameras. Detailed information on each sensor is available in table 2.

Figure 2. M1X

Table 2. Sensor information

BibTeX

@inproceedings{Lee2021IndoorLocDataset,
title={Large-scale Localization Datasets in Crowded Indoor Spaces},
author={Lee, Donghwan and Ryu, Soohyun and Yeon, Suyong and Lee, Yonghan and Kim, Deokhwa and Han, Cheolho and Cabon, Yohann and Weinzaepfel, Philippe and Guérin Nicolas and Csurka, Gabriela and Humenberger, Martin},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year={2021}
}

Go to NAVER LABS Open Dataset page

LIST

LAVER LABS