Namil Kim* (NAVER LAB), Yukyung Choi* (NAVER), Soonmin Hwang (KAIST), In So Kweon (KAIST)
Our multispectral stereo dataset provides the calibrated RGB stereo pair, a co-aligned thermal image with the left-view RGB stereo image and 3D measurements so that the dataset is compatible with various supervised and unsupervised methods. Compared to other multispectral stereo dataset (Barrera, Lumbreras, and Sappa 2013; Treible et al. 2017), we focused on the real-world driving conditions such as campus, residential, urban, and suburb in day and night, and we provide the fully aligned RGB and thermal pairs using the beam- splitter without clipped rectification regions. More specifically, our dataset covers the day [7am to 2pm], night [10pm to 2am], and also the high and under saturated conditions. Totally, we provide (#7383) stereo/thermal or training (#4534) and testing (#2853) in day-time, and testing (#1583) pairs at night.
@INPROCEEDINGS{MTN,
author={N. Kim and Y. Choi and S. Hwang and I. S. Kweon},
booktitle={Association for the Advancement of Artificial Intelligence (AAAI)},
title={Multispectral Transfer Network: Unsupervised Depth Estimation for All-day Vision},
year={2018},
}