DCGAN and CGAN suffer from mode collapse
kunwardeeps opened this issue · 1 comments
kunwardeeps commented
After training for 200 epochs with the CIFAR 10 dataset, I found that the DCGAN and CGAN generated images do not have enough variety. Most likely a mode collapse happening.
mafda commented
Regards,
Mode collapse is one of the main problems that occur during GANs training, which means that the generator produces limited varieties of samples.
In recent work, researchers have presented some studies that suggest some methods to reduce the problem of mode collapse, such as:
- studies that explore new objective functions, like Unrolled GAN and Deep Regret Analytic GAN (DRAGAN),
- studies that propose architectural modifications like Multiagent Diverse GAN (MAD-GAN) and Mode Regularized GAN (MRGAN), and
- studies with mini-batch discrimination like Progressive GAN.
For more information, these works are recommended:
- NIPS 2016 Tutorial: Generative Adversarial Networks
- Generative Adversarial Networks: An Overview
- How Generative Adversarial Networks and Their Variants Work: An Overview
Thanks