Hierarchical Modes Exploring in Generative Adversarial Networks

Authors: Mengxiao Hu, Jinlong Li, Maolin Hu, Tao Hu10981-10988

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validated the proposed algorithm on four conditional image synthesis tasks including categorical generation, paired and un-paired image translation and text-to-image generation. Both qualitative and quantitative results show that the proposed method is effective in alleviating the mode collapse problem in c GANs, and can control the diversity of output images w.r.t specific-level features.
Researcher Affiliation Academia Mengxiao Hu, Jinlong Li, Maolin Hu, Tao Hu University of Science and Technology of China m x hu@126.com, jlli@ustc.edu.cn, {humaolin, Skyful}@mail.ustc.edu.cn
Pseudocode No The paper does not contain pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any statement or link for open-source code availability.
Open Datasets Yes Categorical generation, it is trained on CIFAR-10 (Szegedy et al. 2015)... Paired image-to-image translation, it is trained on facades and maps using Pix2Pix as the baseline model. Unpaired image-to-image translation, it is trained on Yosemite (Zhu et al. 2017a) and cat dog (Lee et al. 2018)... Text-to-image generation, it is trained on CUB-200-2011 (Wah et al. 2011)...
Dataset Splits No The paper does not explicitly provide specific percentages or sample counts for training, validation, and test splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes Because the original networks of the baseline model do not change after adding the attention unit and the regularization term, we kept the hyper-parameters of the baseline model original. We adopted L1 norm as distance metrics for all d(i)( ) and set the weight of regularization β = 1 in all experiments.