Generative Adversarial Networks (GANs) are powerful models able to synthesize data samples closely resembling the distribution of real data, yet the diversity of those generated samples is limited due to the so-called mode collapse phenomenon observed in GANs. Conditional GANs are especially prone to mode collapse, as they tend to ignore the input noise vector and focus on the conditional information. Recent methods proposed to mitigate this limitation increase the diversity of generated samples, yet they reduce the performance of the models when similarity of samples is required. To address this shortcoming, we propose a novel method to control the diversity of GAN-generated samples. By adding a simple, yet effective regularization to the training loss function we encourage the generator to discover new data modes for inputs related to diverse outputs while generating consistent samples for the remaining ones. More precisely, we reward or penalize the model for synthesising diverse images, matching the diversity of real and generated samples for a given conditional input. We show the superiority of our method on simulating data from the Zero Degree Calorimeter of the ALICE experiment in LHC, CERN.
Jan Dubiński (Warsaw University of Technology)
Kamil Deja (Institute of Computer Science, Warsaw University of Technology, Nowowiejska 15/19 00-665 Warszawa, NIP: 5250005834)
Sandro Wenzel (CERN)
Przemysław Rokita (Warsaw University of Technology)
Przemysław ROKITA PhD DSc is a tenured professor at the Warsaw University of Technology and head of the Computer Graphics Division. He is a member of SPIE, ACM, IEEE and Committee on Informatics of the Polish Academy of Sciences. His research interests include: computer science & information technology, digital image processing, computer graphics, image perception, computer vision, artificial intelligence, data analysis, virtual reality, computer games, computer engineering, human-computer interaction; Previously affiliated as visiting scientist and professor at: Max-Planck-Institut für Informatik - Computer Graphics Department (Germany), The University of Aizu (Japan), Hiroshima Institute of Technology (Japan), Hiroshima Prefectural University (Japan), Imperial College of Science, Technology and Medicine (United Kingdom); Member of Program Committees and reviewer for many international conferences and journals, including: IEEE Computer Graphics & Applications, The Visual Computer, Real-Time Imaging, Opto-Electronics Review, Journal of Imaging Science and Technology, IEEE Transactions on Circuits and Systems for Video Technology, IEEE Transactions on Multimedia, ACM Siggraph, Eurographics, High Performance Graphics. Expert and consultant at the National Centre for Research and Development, National Science Centre, Ministry of Science and Higher Education.
Tomasz Trzcinski (Warsaw University of Technology, Tooploox, IDEAS, Jagiellonian University)
More from the Same Authors
2022 Poster: FlowHMM: Flow-based continuous hidden Markov models »
Pawel Lorek · Rafal Nowak · Tomasz Trzcinski · Maciej Zieba
2022 Poster: On Analyzing Generative and Denoising Capabilities of Diffusion-based Deep Generative Models »
Kamil Deja · Anna Kuzina · Tomasz Trzcinski · Jakub Tomczak