Bidirectional Conditional Generative Adversarial Networks
By Ayush Jaiswal, Wael AbdAlmageed, Yue Wu, Premkumar Natarajan. 2017
42 citations recorded by the source corpus.
Conditional Generative Adversarial Networks (cGANs) are generative models that can produce data samples ($x$) conditioned on both latent variables ($z$) and known auxiliary information ($c$). We propose the Bidirectional cGAN (BiCoGAN), which effectively disentangles $z$ and $c$ in the generation process and provides an encoder that learns inverse mappings from $x$ to both $z$ and $c$, trained jointly with the generator and the discriminator. We present crucial techniques for training BiCoGANs, which involve an extrinsic factor loss along with an associated dynamically-tuned importance weight. As compared to other encoder-based cGANs, BiCoGANs encode $c$ more accurately, and utilize $z$ and $c$ more effectively and in a more disentangled way to generate samples.
Subfields: Computer Vision and Pattern Recognition
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Bernhard Schölkopf
- Luc Van Gool
- Andrew Zisserman
- Dacheng Tao
- Xiaogang Wang
- Andrea Vedaldi
- Trevor Darrell
- Bernt Schiele
Citations in the corpus
- Sample-Efficient Reinforcement Learning via Counterfactual-Based Data Augmentation (2020, methodology): “To achieve the two goals, estimating f and noise values simultaneously, we cast the learning of both an inference machine (encoder) and a deep generative model (decoder) in a generative adversarial network (GAN)-like adversarial framework, called Bidirectional Conditional GAN (BiCoGAN [16]).”
- CapsuleGAN: Generative Adversarial Capsule Network (2018, methodology): “The results presented in this paper motivate the use of CapsNets as opposed to CNNs for encoders in GAN variants like BiCoGAN [11].”
- Generative Adversarial Networks (GANs) (2020, background): “Re-engineered Network Architecture (S1) Conditional generation (S11) cGANs [34], FCGAN [53], IRGAN [53], GRANs [54], LAPGAN [5], SGAN [49], IcGAN [55], BiCoGAN [56], MatAN [57], Self-conditioned GANs [58], AC-GANs [59], TripleGAN [60], KDGAN [61], ControlGAN [62]”
- Guided Image Generation with Conditional Invertible Neural Networks (2019, background): “This is avoided by AGE networks [39] and Cy-cleGANs [43], which combine standard cGAN discrimi-nators with L1 reconstruction loss in the data domain, and bidirectional conditional GANs [20], which extend the GAN discriminator to act on the distributions in data and latent space jointly.”