Our dataset is available for academic research. To acquire this dataset, please download this agreement, read it carefully and send it back to us with the signature of the director of your school or your institution. For copyright protection, we only accept the application from the director of your school or your institution.
Our dataset includes the facial scans of 438 subjects. For each subject, we capture 20 expressions. The synthetic relighting dataset is rendered using the scans of 100 subjects, which includes 270,000 valid image groups for training and 20,00 valid image groups for testing.
Some data is still under the examinations in SenseTime. Currently, we can provide the scans of 270 subjects. The relighting dataset includes the images of 80 subjects.
As the agreement is based on the entire dataset, you do not need to sign the agreement again for the data released in the future.
This work was supported by the National Key R&D Program of China 2018YFA0704000, the NSFC (No.61822111, 61727808, 61671268) and Beijing Natural Science Foundation (JQ19015, L182052). Feng Xu is the corresponding author. The 3rd author (Ming Lu) waives his intellectual property rights of this work. We thank Tiancheng Sun for providing the relit results of the real images.
@article{10.1145/3414685.3417824,
author = {Wang, Zhibo and Yu, Xin and Lu, Ming and Wang, Quan and Qian, Chen and Xu, Feng},
title = {Single Image Portrait Relighting via Explicit Multiple Reflectance Channel Modeling},
year = {2020},
issue_date = {December 2020},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {39},
number = {6},
issn = {0730-0301},
doi = {10.1145/3414685.3417824},
journal = {ACM Trans. Graph.},
month = nov,
articleno = {220},
numpages = {13},
}