Improved Consistency Regularization for GANs
Authors: Zhengli Zhao, Sameer Singh, Honglak Lee, Zizhao Zhang, Augustus Odena, Han Zhang11033-11041
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We carry out extensive experiments quantifying the benefit of our improvements. For unconditional image synthesis on CIFAR-10 and Celeb A, our modifications yield the best known FID scores on various GAN architectures. For conditional image synthesis on CIFAR-10, we improve the state-of-the-art FID score from 11.48 to 9.21. Finally, on Image Net-2012, we apply our technique to the original Big GAN model and improve the FID from 6.66 to 5.38, which is the best score at that model size. |
| Researcher Affiliation | Collaboration | Zhengli Zhao,1,2 Sameer Singh,1 Honglak Lee,2 Zizhao Zhang,2 Augustus Odena,2 Han Zhang2 1 University of California, Irvine 2 Google Research |
| Pseudocode | Yes | Algorithm 1 Balanced Consistency Regularization (b CR) ... Algorithm 2 Latent Consistency Regularization (z CR) |
| Open Source Code | No | The paper does not provide an explicit statement or link indicating that the source code for their methodology is publicly available. |
| Open Datasets | Yes | We carry out extensive experiments comparing our methods against the above baselines on three commonly used data sets in the GAN literature: CIFAR-10 (Krizhevsky, Hinton et al. 2009), Celeb A-HQ-128 (Karras et al. 2018), and Image Net2012 (Russakovsky et al. 2015). |
| Dataset Splits | No | The paper specifies training and testing splits for CIFAR-10 (50K training, 10K testing) and Celeb A (train models with the rest, 3K testing), but it does not explicitly mention a distinct validation set or its size for hyperparameter tuning or early stopping during training. While it mentions reporting the FID distribution of the top 15% of trained models, this doesn't constitute a described validation split. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for experiments, such as GPU models, CPU types, or cloud computing instance specifications. |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies used in the experiments. |
| Experiment Setup | Yes | For optimization, we use the Adam optimizer with batch size of 64 for all experiments. By default, spectral normalization (SN) (Miyato et al. 2018a) is used in the discriminator... We stop training after 200k generator update steps for CIFAR-10, 100k steps for Celeb A, and 250k for Image Net. We set λreal = λfake = 10 for b CR, while using σnoise = 0.03, λgen = 0.5, and λdis = 5 for z CR. |