{
  "id": 279028,
  "title": "using GAN (generative adversarial net) ...",
  "url": "/competitions/sartorius-cell-instance-segmentation/discussion/279028",
  "author_name": "",
  "post_date": "2021-10-16T13:28:36.869136400Z",
  "votes": 9,
  "comment_count": 1,
  "views": 0,
  "content": "<p>For those who are fans of GAN, you can try this. Using GAN for domain transfer in semi-supervised setting:</p>\n<p><img src=\"https://www.linkpicture.com/q/dsdfsdfdsf.png\" alt=\"\"></p>\n<p>There are many codes available. But you have to modify it for the segmentation case.<br>\nA word of warning: GAN are generally not easy to work with … you are balancing two losses and try to find the saddle point. You don't get good results unless you are experienced to adjust the learning parameters</p>\n<p>Related papers:</p>\n<p>(1) \"Unsupervised Domain Adaptation by Backpropagation\" - Yaroslav Ganin, Victor Lempitsky, ARXIV 2014</p>\n<p>(2) \"Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning\" - Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, Shin Ishii, ARXIV 2017</p>\n<p><a href=\"https://github.com/9310gaurav/virtual-adversarial-training\" target=\"_blank\">https://github.com/9310gaurav/virtual-adversarial-training</a></p>",
  "messages": [
    {
      "id": "1546740",
      "postDate": "10/16/2021 13:28:36",
      "content": "<p>For those who are fans of GAN, you can try this. Using GAN for domain transfer in semi-supervised setting:</p>\n<p><img src=\"https://www.linkpicture.com/q/dsdfsdfdsf.png\" alt=\"\"></p>\n<p>There are many codes available. But you have to modify it for the segmentation case.<br>\nA word of warning: GAN are generally not easy to work with … you are balancing two losses and try to find the saddle point. You don't get good results unless you are experienced to adjust the learning parameters</p>\n<p>Related papers:</p>\n<p>(1) \"Unsupervised Domain Adaptation by Backpropagation\" - Yaroslav Ganin, Victor Lempitsky, ARXIV 2014</p>\n<p>(2) \"Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning\" - Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, Shin Ishii, ARXIV 2017</p>\n<p><a href=\"https://github.com/9310gaurav/virtual-adversarial-training\" target=\"_blank\">https://github.com/9310gaurav/virtual-adversarial-training</a></p>",
      "rawMarkdown": "For those who are fans of GAN, you can try this. Using GAN for domain transfer in semi-supervised setting:\n\n![](https://www.linkpicture.com/q/dsdfsdfdsf.png)\n\nThere are many codes available. But you have to modify it for the segmentation case.\nA word of warning: GAN are generally not easy to work with … you are balancing two losses and try to find the saddle point. You don't get good results unless you are experienced to adjust the learning parameters\n\nRelated papers:\n\n(1) \"Unsupervised Domain Adaptation by Backpropagation\" - Yaroslav Ganin, Victor Lempitsky, ARXIV 2014\n\n(2) \"Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning\" - Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, Shin Ishii, ARXIV 2017\n\n[https://github.com/9310gaurav/virtual-adversarial-training](https://github.com/9310gaurav/virtual-adversarial-training)",
      "votes": null
    },
    {
      "id": "1572877",
      "postDate": "11/06/2021 03:43:11",
      "content": "<p>Thank you.  I am glad to know how to use unlabeled data.  I will try this.</p>",
      "rawMarkdown": "Thank you.  I am glad to know how to use unlabeled data.  I will try this.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1572877,
      "author_name": "aikaro",
      "author_url": "",
      "post_date": "11/06/2021 03:43:11",
      "content": "<p>Thank you.  I am glad to know how to use unlabeled data.  I will try this.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1546740": "For those who are fans of GAN, you can try this. Using GAN for domain transfer in semi-supervised setting:\n\n![](https://www.linkpicture.com/q/dsdfsdfdsf.png)\n\nThere are many codes available. But you have to modify it for the segmentation case.\nA word of warning: GAN are generally not easy to work with … you are balancing two losses and try to find the saddle point. You don't get good results unless you are experienced to adjust the learning parameters\n\nRelated papers:\n\n(1) \"Unsupervised Domain Adaptation by Backpropagation\" - Yaroslav Ganin, Victor Lempitsky, ARXIV 2014\n\n(2) \"Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning\" - Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, Shin Ishii, ARXIV 2017\n\n[https://github.com/9310gaurav/virtual-adversarial-training](https://github.com/9310gaurav/virtual-adversarial-training)",
    "1572877": "Thank you.  I am glad to know how to use unlabeled data.  I will try this."
  },
  "source": "meta"
}