Spectral Distribution Aware Image Generation

Authors: Steffen Jung, Margret Keuper1734-1742

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments are conducted on the FFHQ dataset, which contains 70 000 high-quality face images at 1024 1024 resolution showing large variations in terms of age, ethnicity, and backgrounds. From this dataset, we downsample three versions in resolution 64 64, 128 128 and 256 256. We evaluate all experiments on 10k examples in terms of FID and the proposed cloaking score (CS).
Researcher Affiliation Academia 1 Max Planck Institute for Informatics, Saarland Informatics Campus 2 Data and Web Science Group, University of Mannheim
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Our implementation is publicly available1. 1https://github.com/steffen-jung/Spectral GAN
Open Datasets Yes Experiments are conducted on the FFHQ dataset, which contains 70 000 high-quality face images at 1024 1024 resolution showing large variations in terms of age, ethnicity, and backgrounds. From this dataset, we downsample three versions in resolution 64 64, 128 128 and 256 256.
Dataset Splits No The paper mentions training epochs and evaluation, but does not provide specific training, validation, or test split percentages or counts for the datasets used.
Hardware Specification No The paper does not mention any specific hardware specifications used for running the experiments.
Software Dependencies No The paper mentions "Adam optimizer" and "RMSprop" but does not specify version numbers for any software dependencies or libraries.
Experiment Setup Yes DCGAN, LSGAN and WGAN-GP were trained using Adam optimizer, for WGAN, we used RMSprop, with a learning rate of 0.0002 and a batch size of 128 for 500 epochs. In all cases, we apply the same loss function on both discriminators.