Litterature
GAN background
- Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014.
- Salimans, Tim, et al. "Improved techniques for training gans." Advances in Neural Information Processing Systems. 2016.
- Arjovsky, Martin, Soumith Chintala, and Léon Bottou. "Wasserstein gan." arXiv preprint arXiv:1701.07875 (2017).
- Mirza, Mehdi, and Simon Osindero. "Conditional generative adversarial nets." arXiv preprint arXiv:1411.1784 (2014).
- Lucic, Mario, et al. "Are GANs Created Equal? A Large-Scale Study." arXiv preprint arXiv:1711.10337 (2017).
Image-to-image models
- Isola, Phillip, et al. "Image-to-image translation with conditional adversarial networks." arXiv preprint arXiv:1611.07004 (2016).
- Zhu, Jun-Yan, et al. "Unpaired image-to-image translation using cycle-consistent adversarial networks." arXiv preprint arXiv:1703.10593 (2017).
- Shrivastava, Ashish, et al. "Learning from simulated and unsupervised images through adversarial training." arXiv preprint arXiv:1612.07828 (2016).
- Zhu, Jun-Yan, et al. "Toward Multimodal Image-to-Image Translation." arXiv preprint arXiv:1711.11586 (2017).
Interesting ideas
Promising results
- Karras, Tero, et al. "Progressive growing of gans for improved quality, stability, and variation." arXiv preprint arXiv:1710.10196 (2017).
- Miyato, Takeru, et al. "Spectral Normalization for Generative Adversarial Networks."
Project plan
Right now I have no better place to write this so I'll do it here. Due to the long time consumption of the experiments I propose focusing on two things with two corresponding large-scale experiments. Progressive growing and triple-GAN. Firstly i will design a new training framework with these constellations in mind, thereafter I will test how well the progressive-growing works. When this have been tested I will focus on smaller rapid experiments on low-resolution images for a while to manage randomness and variation as well as getting the triple-GAN constellation to work properly. When this have beem performed I will scale up for a large experiment resulting in a high-resolution GAN.