Selected article for: "high performance and model performance"
Author: Huang, Gaofeng; Jafari, Amir Hossein
Title: Enhanced balancing GAN: minority-class image generation
  • Cord-id: pkympgsc
  • Document date: 2021_6_17
    • ID: pkympgsc
    Snippet: Generative adversarial networks (GANs) are one of the most powerful generative models, but always require a large and balanced dataset to train. Traditional GANs are not applicable to generate minority-class images in a highly imbalanced dataset. Balancing GAN (BAGAN) is proposed to mitigate this problem, but it is unstable when images in different classes look similar, e.g., flowers and cells. In this work, we propose a supervised autoencoder with an intermediate embedding model to disperse the
    Document: Generative adversarial networks (GANs) are one of the most powerful generative models, but always require a large and balanced dataset to train. Traditional GANs are not applicable to generate minority-class images in a highly imbalanced dataset. Balancing GAN (BAGAN) is proposed to mitigate this problem, but it is unstable when images in different classes look similar, e.g., flowers and cells. In this work, we propose a supervised autoencoder with an intermediate embedding model to disperse the labeled latent vectors. With the enhanced autoencoder initialization, we also build an architecture of BAGAN with gradient penalty (BAGAN-GP). Our proposed model overcomes the unstable issue in original BAGAN and converges faster to high-quality generations. Our model achieves high performance on the imbalanced scale-down version of MNIST Fashion, CIFAR-10, and one small-scale medical image dataset. https://github.com/GH920/improved-bagan-gp.

    Search related documents:
    Co phrase search for related documents
    • activation function and loss function: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
    • adam algorithm and loss function: 1, 2
    • long training and loss sensitive: 1