Progressive Growing of GANs for Improved Quality, Stability, and Variation
Authors: Tero Karras, Timo Aila, Samuli Laine, Jaakko Lehtinen
ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our contributions using the CELEBA, LSUN, CIFAR10 datasets. We improve the best published inception score for CIFAR10. |
| Researcher Affiliation | Collaboration | Tero Karras NVIDIA {tkarras,taila,slaine,jlehtinen}@nvidia.com Timo Aila NVIDIA Samuli Laine NVIDIA Jaakko Lehtinen NVIDIA and Aalto University |
| Pseudocode | No | The paper describes methods and network architectures (Table 2) but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | This dataset and our full implementation are available at https://github.com/tkarras/progressive_growing_of_gans, trained networks can be found at https://drive.google.com/open?id=0B4qLcYyJmiz0NHFULTdYc05lX0U along with result images, and a supplementary video illustrating the datasets, additional results, and latent space interpolations is at https://youtu.be/G06dEcZ-QTg. |
| Open Datasets | Yes | We evaluate our contributions using the CELEBA, LSUN, CIFAR10 datasets. We have also created a higher quality version of the CELEBA dataset that allows experimentation with output resolutions up to 1024x1024 pixels. This dataset and our full implementation are available at https://github.com/tkarras/progressive_growing_of_gans |
| Dataset Splits | No | The paper mentions using CELEBA, LSUN, and CIFAR10 datasets but does not explicitly state specific training, validation, or test dataset splits (e.g., percentages or counts) for reproduction. It mentions "validation" in the context of evaluating metrics, not data partitioning. |
| Hardware Specification | Yes | We trained the network on 8 Tesla V100 GPUs for 4 days, after which we no longer observed qualitative differences between the results of consecutive training iterations. |
| Software Dependencies | No | The paper mentions using "Adam" optimizer and "WGAN-GP loss" but does not provide specific version numbers for any software libraries or dependencies (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | We train the networks using Adam (Kingma & Ba, 2015) with α = 0.001, β1 = 0, β2 = 0.99, and ϵ = 10^-8. We use a minibatch size 16 for resolutions 4x4 - 128x128 and then gradually decrease the size according to 256x256 - 14, 512x512 - 6, and 1024x1024 - 3 to avoid exceeding the available memory budget. We use the WGAN-GP loss, but unlike Gulrajani et al. (2017), we alternate between optimizing the generator and discriminator on a per-minibatch basis, i.e., we set ncritic = 1. Additionally, we introduce a fourth term into the discriminator loss with an extremely small weight to keep the discriminator output from drifting too far away from zero. To be precise, we set L = L + ϵdrift Ex Pr[D(x)^2], where ϵdrift = 0.001. |