Sanghun Kim
ABSTRACT
Recently Top-k fake sample selection has been introduced to provide better gradients for training Generative Adversarial Networks. Since the method does not guarantee class balance of selected samples in class conditional GANs, certain classes can be completely ignored in the training. In this work, we propose class standardized critic score based sample selection which enables class balanced sample selection. Our method achieves improved FID score and Intra-FID score compared to prior Top-k selection.
Supplemental Material
- Andrew Brock, Jeff Donahue, and Karen Simonyan. 2018. Large Scale GAN Training for High Fidelity Natural Image Synthesis. International Conference on Learning Representations (2018).Google Scholar
- Terrance DeVries, Michal Drozdzal, and Graham W. Taylor. 2020. Instance Selection for GANs. NIPS (2020).Google Scholar
- Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. 2017. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium. NIPS (2017).Google Scholar
- Samarth Sinha, Zhengli Zhao, Anirudh Goyal, Colin Raffel, and Augustus Odena. 2020. Top-k Training of GANs: Improving GAN Performance by Throwing Away Bad Samples. NIPS (2020).Google Scholar
Recommendations
Subspace Distance-Based Sampling Method for SVM
ICDMW '10: Proceedings of the 2010 IEEE International Conference on Data Mining WorkshopsSupport Vector Machine (SVM) is an effective classifier for classification task, but a vital shortcoming of SVM is that it needs huge computation for large scale learning tasks. Sample selection is a feasible strategy to overcome the problem. In order ...
Simplifying Bias Correction for Selective Sampling: A Unified Distribution-Free Approach to Handling Endogenously Selected Samples
Exploiting informative endogenously selected samples while minimizing sample selection bias with minimum modeling assumptions and analytical burden.
Unlike random sampling, selective sampling draws units based on the outcome values, such as oversampling rare events in choice outcomes and extreme activities on continuous and count outcomes. Despite high cost-effectiveness for marketing research, such ...
Fast balanced sampling for highly stratified population
Balanced sampling is a very efficient sampling design when the variable of interest is correlated to the auxiliary variables on which the sample is balanced. A procedure to select balanced samples in a stratified population has previously been proposed. ...
Comments