On the Analysis of GAN-based Image-to-Image Translation with Gaussian Noise Injection

Authors: Chaohua Shi, Kexin Huang, Lu GAN, Hongqing Liu, Mingrui Zhu, Nannan Wang, Xinbo Gao

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments validate our theoretical findings, showing substantial improvements over various I2I baseline models in noisy settings.
Researcher Affiliation Academia Chaohua Shi1 Kexin Huang2 Lu Gan ,3 Hongqing Liu4 Mingrui Zhu1 Nannan Wang ,1 Xinbo Gao1,4 1 Xidian University 2 National University of Defense Technology 3 Brunel University 4 Chongqing University of Posts and Telecommunications
Pseudocode No The paper describes mathematical derivations and experimental procedures but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper lists links to baseline models used (e.g., 'https://github.com/Lotayou/Face-Renovation', 'https://github.com/williamyang1991/GP-UNIT') and mentions 'shared by the authors' for another, but does not explicitly state that the source code for their proposed Gaussian Noise Injection methodology is open-sourced or provide a link to it.
Open Datasets Yes During training, we follow the default settings of each baseline model and use the corresponding datasets: FFHQ (Karras et al., 2019) (Hi Face GAN), AAHQ (Choi et al., 2020) (GP-UNIT), and CUFS (Wang et al., 2018) (Sketch Transformer).
Dataset Splits No The paper specifies training and testing subsets for FFHQ ('10000 images... for training, while the final 1000 images were used for testing') and mentions test images for AFHQ ('Each domain comprises 500 test images'). However, it does not explicitly detail a separate validation split or the precise splits for all datasets used, making full reproduction of the data partitioning unclear without further assumptions.
Hardware Specification No The paper does not specify the hardware used for experiments, such as particular GPU or CPU models, memory configurations, or cloud computing resources.
Software Dependencies No The paper mentions various models and evaluation metrics (e.g., GP-UNIT, Sketch Transformer, FID, LPIPS) and external tools (e.g., CBM3D), but it does not specify any software versions for these or for the underlying programming languages or libraries used in the implementation.
Experiment Setup Yes During training, zero-mean, isotropic Gaussian noise with σ2 t = 0.04 (unless specified otherwise) is introduced.