TY - GEN
T1 - Unsupervised Holistic Image Generation from Key Local Patches
AU - Lee, Donghoon
AU - Yun, Sangdoo
AU - Choi, Sungjoon
AU - Yoo, Hwiyeon
AU - Yang, Ming Hsuan
AU - Oh, Songhwai
N1 - Funding Information:
The work of D. Lee, S. Choi, H. Yoo, and S. Oh is supported in part by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (NRF-2017R1A2B2006136) and by ‘The Cross-Ministry Giga KOREA Project’ grant funded by the Korea gov-ernment(MSIT) (No. GK18P0300, Real-time 4D reconstruction of dynamic objects for ultra-realistic service). The work of M.-H. Yang is supported in part by the National Natural Science Foundation of China under Grant #61771288, the NSF CAREER Grant #1149783, and gifts from Adobe and Nvidia.
Publisher Copyright:
© 2018, Springer Nature Switzerland AG.
PY - 2018
Y1 - 2018
N2 - We introduce a new problem of generating an image based on a small number of key local patches without any geometric prior. In this work, key local patches are defined as informative regions of the target object or scene. This is a challenging problem since it requires generating realistic images and predicting locations of parts at the same time. We construct adversarial networks to tackle this problem. A generator network generates a fake image as well as a mask based on the encoder-decoder framework. On the other hand, a discriminator network aims to detect fake images. The network is trained with three losses to consider spatial, appearance, and adversarial information. The spatial loss determines whether the locations of predicted parts are correct. Input patches are restored in the output image without much modification due to the appearance loss. The adversarial loss ensures output images are realistic. The proposed network is trained without supervisory signals since no labels of key parts are required. Experimental results on seven datasets demonstrate that the proposed algorithm performs favorably on challenging objects and scenes.
AB - We introduce a new problem of generating an image based on a small number of key local patches without any geometric prior. In this work, key local patches are defined as informative regions of the target object or scene. This is a challenging problem since it requires generating realistic images and predicting locations of parts at the same time. We construct adversarial networks to tackle this problem. A generator network generates a fake image as well as a mask based on the encoder-decoder framework. On the other hand, a discriminator network aims to detect fake images. The network is trained with three losses to consider spatial, appearance, and adversarial information. The spatial loss determines whether the locations of predicted parts are correct. Input patches are restored in the output image without much modification due to the appearance loss. The adversarial loss ensures output images are realistic. The proposed network is trained without supervisory signals since no labels of key parts are required. Experimental results on seven datasets demonstrate that the proposed algorithm performs favorably on challenging objects and scenes.
KW - Generative adversarial networks
KW - Image synthesis
UR - http://www.scopus.com/inward/record.url?scp=85055084000&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-01228-1_2
DO - 10.1007/978-3-030-01228-1_2
M3 - Conference contribution
AN - SCOPUS:85055084000
SN - 9783030012274
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 21
EP - 37
BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
A2 - Ferrari, Vittorio
A2 - Sminchisescu, Cristian
A2 - Hebert, Martial
A2 - Weiss, Yair
PB - Springer Verlag
T2 - 15th European Conference on Computer Vision, ECCV 2018
Y2 - 8 September 2018 through 14 September 2018
ER -