Metric-based generative adversarial network

Guoxian Dai, Jin Xie, Yi Fang

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    Existing methods of generative adversarial network (GAN) use different criteria to distinguish between real and fake samples, such as probability [9], energy [44] or other losses [30]. In this paper, by employing the merits of deep metric learning, we propose a novel metric-based generative adversarial network (MBGAN), which uses the distance-criteria to distinguish between real and fake samples. Specifically, the discriminator of MBGAN adopts a triplet structure and learns a deep nonlinear transformation, which maps input samples into a new feature space. In the transformed space, the distance between real samples is minimized, while the distance between real sample and fake sample is maximized. Similar to the adversarial procedure of existing GANs, a generator is trained to produce synthesized examples, which are close to real examples, while a discriminator is trained to maximize the distance between real and fake samples to a large margin. Meanwhile, instead of using a fixed margin, we adopt a data-dependent margin [30], so that the generator could focus on improving the synthesized samples with poor quality, instead of wasting energy on well-produce samples. Our proposed method is verified on various benchmarks, such as CIFAR-10, SVHN and CelebA, and generates high-quality samples.

    Original languageEnglish (US)
    Title of host publicationMM 2017 - Proceedings of the 2017 ACM Multimedia Conference
    PublisherAssociation for Computing Machinery, Inc
    Pages672-680
    Number of pages9
    ISBN (Electronic)9781450349062
    DOIs
    StatePublished - Oct 23 2017
    Event25th ACM International Conference on Multimedia, MM 2017 - Mountain View, United States
    Duration: Oct 23 2017Oct 27 2017

    Other

    Other25th ACM International Conference on Multimedia, MM 2017
    CountryUnited States
    CityMountain View
    Period10/23/1710/27/17

    Fingerprint

    Discriminators

    Keywords

    • Data-dependent margin
    • Deep metric learning
    • Generative adversarial network

    ASJC Scopus subject areas

    • Computer Graphics and Computer-Aided Design
    • Media Technology
    • Computer Vision and Pattern Recognition
    • Software

    Cite this

    Dai, G., Xie, J., & Fang, Y. (2017). Metric-based generative adversarial network. In MM 2017 - Proceedings of the 2017 ACM Multimedia Conference (pp. 672-680). Association for Computing Machinery, Inc. https://doi.org/10.1145/3123266.3123334

    Metric-based generative adversarial network. / Dai, Guoxian; Xie, Jin; Fang, Yi.

    MM 2017 - Proceedings of the 2017 ACM Multimedia Conference. Association for Computing Machinery, Inc, 2017. p. 672-680.

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Dai, G, Xie, J & Fang, Y 2017, Metric-based generative adversarial network. in MM 2017 - Proceedings of the 2017 ACM Multimedia Conference. Association for Computing Machinery, Inc, pp. 672-680, 25th ACM International Conference on Multimedia, MM 2017, Mountain View, United States, 10/23/17. https://doi.org/10.1145/3123266.3123334
    Dai G, Xie J, Fang Y. Metric-based generative adversarial network. In MM 2017 - Proceedings of the 2017 ACM Multimedia Conference. Association for Computing Machinery, Inc. 2017. p. 672-680 https://doi.org/10.1145/3123266.3123334
    Dai, Guoxian ; Xie, Jin ; Fang, Yi. / Metric-based generative adversarial network. MM 2017 - Proceedings of the 2017 ACM Multimedia Conference. Association for Computing Machinery, Inc, 2017. pp. 672-680
    @inproceedings{397ad330289f4eb29e794fa981103bfd,
    title = "Metric-based generative adversarial network",
    abstract = "Existing methods of generative adversarial network (GAN) use different criteria to distinguish between real and fake samples, such as probability [9], energy [44] or other losses [30]. In this paper, by employing the merits of deep metric learning, we propose a novel metric-based generative adversarial network (MBGAN), which uses the distance-criteria to distinguish between real and fake samples. Specifically, the discriminator of MBGAN adopts a triplet structure and learns a deep nonlinear transformation, which maps input samples into a new feature space. In the transformed space, the distance between real samples is minimized, while the distance between real sample and fake sample is maximized. Similar to the adversarial procedure of existing GANs, a generator is trained to produce synthesized examples, which are close to real examples, while a discriminator is trained to maximize the distance between real and fake samples to a large margin. Meanwhile, instead of using a fixed margin, we adopt a data-dependent margin [30], so that the generator could focus on improving the synthesized samples with poor quality, instead of wasting energy on well-produce samples. Our proposed method is verified on various benchmarks, such as CIFAR-10, SVHN and CelebA, and generates high-quality samples.",
    keywords = "Data-dependent margin, Deep metric learning, Generative adversarial network",
    author = "Guoxian Dai and Jin Xie and Yi Fang",
    year = "2017",
    month = "10",
    day = "23",
    doi = "10.1145/3123266.3123334",
    language = "English (US)",
    pages = "672--680",
    booktitle = "MM 2017 - Proceedings of the 2017 ACM Multimedia Conference",
    publisher = "Association for Computing Machinery, Inc",

    }

    TY - GEN

    T1 - Metric-based generative adversarial network

    AU - Dai, Guoxian

    AU - Xie, Jin

    AU - Fang, Yi

    PY - 2017/10/23

    Y1 - 2017/10/23

    N2 - Existing methods of generative adversarial network (GAN) use different criteria to distinguish between real and fake samples, such as probability [9], energy [44] or other losses [30]. In this paper, by employing the merits of deep metric learning, we propose a novel metric-based generative adversarial network (MBGAN), which uses the distance-criteria to distinguish between real and fake samples. Specifically, the discriminator of MBGAN adopts a triplet structure and learns a deep nonlinear transformation, which maps input samples into a new feature space. In the transformed space, the distance between real samples is minimized, while the distance between real sample and fake sample is maximized. Similar to the adversarial procedure of existing GANs, a generator is trained to produce synthesized examples, which are close to real examples, while a discriminator is trained to maximize the distance between real and fake samples to a large margin. Meanwhile, instead of using a fixed margin, we adopt a data-dependent margin [30], so that the generator could focus on improving the synthesized samples with poor quality, instead of wasting energy on well-produce samples. Our proposed method is verified on various benchmarks, such as CIFAR-10, SVHN and CelebA, and generates high-quality samples.

    AB - Existing methods of generative adversarial network (GAN) use different criteria to distinguish between real and fake samples, such as probability [9], energy [44] or other losses [30]. In this paper, by employing the merits of deep metric learning, we propose a novel metric-based generative adversarial network (MBGAN), which uses the distance-criteria to distinguish between real and fake samples. Specifically, the discriminator of MBGAN adopts a triplet structure and learns a deep nonlinear transformation, which maps input samples into a new feature space. In the transformed space, the distance between real samples is minimized, while the distance between real sample and fake sample is maximized. Similar to the adversarial procedure of existing GANs, a generator is trained to produce synthesized examples, which are close to real examples, while a discriminator is trained to maximize the distance between real and fake samples to a large margin. Meanwhile, instead of using a fixed margin, we adopt a data-dependent margin [30], so that the generator could focus on improving the synthesized samples with poor quality, instead of wasting energy on well-produce samples. Our proposed method is verified on various benchmarks, such as CIFAR-10, SVHN and CelebA, and generates high-quality samples.

    KW - Data-dependent margin

    KW - Deep metric learning

    KW - Generative adversarial network

    UR - http://www.scopus.com/inward/record.url?scp=85035208332&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85035208332&partnerID=8YFLogxK

    U2 - 10.1145/3123266.3123334

    DO - 10.1145/3123266.3123334

    M3 - Conference contribution

    SP - 672

    EP - 680

    BT - MM 2017 - Proceedings of the 2017 ACM Multimedia Conference

    PB - Association for Computing Machinery, Inc

    ER -