{
  "id": 125012,
  "title": "Does DFDC do the opposite of its objective?",
  "url": "/competitions/deepfake-detection-challenge/discussion/125012",
  "author_name": "",
  "post_date": "2020-01-08T05:03:00.661367900Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello everyone! I wanted to share a thought with you all. It may be flawed, so take it with a grain of salt.</p>\n\n<p>DFDC challenges participants to build an accurate model to identify deep fake videos. So, in GAN terminology, we are building a \"discriminator\". Based on my understanding, GANs (which are used to generate fake videos) use two networks, a \"discriminator\" and a \"generator\" that compete in a zero-sum game. If we build and open-source better and better \"discriminator\" models, deep fake creators can leverage these models to build better and better \"generators\". Wouldn't that defeat the purpose of the competition by catalyzing the growth of powerful generative models?</p>\n\n<p>This might result in a chain reaction where researchers keep open-sourcing better \"discriminator\" models and deep fake creators use these models to build better \"generative\" models. <strong>So, in a way, the researchers and deep fake creators are unintentionally collaborating.</strong></p>\n\n<p>Please share your opinion on this! I know this may not make any sense, so please correct me if I am wrong :)</p>\n\n<p>(Credit for the idea: <a href=\"/pranavpulijala\">@pranavpulijala</a>)</p>",
  "messages": [
    {
      "id": "713268",
      "postDate": "01/08/2020 05:03:00",
      "content": "<p>Hello everyone! I wanted to share a thought with you all. It may be flawed, so take it with a grain of salt.</p>\n\n<p>DFDC challenges participants to build an accurate model to identify deep fake videos. So, in GAN terminology, we are building a \"discriminator\". Based on my understanding, GANs (which are used to generate fake videos) use two networks, a \"discriminator\" and a \"generator\" that compete in a zero-sum game. If we build and open-source better and better \"discriminator\" models, deep fake creators can leverage these models to build better and better \"generators\". Wouldn't that defeat the purpose of the competition by catalyzing the growth of powerful generative models?</p>\n\n<p>This might result in a chain reaction where researchers keep open-sourcing better \"discriminator\" models and deep fake creators use these models to build better \"generative\" models. <strong>So, in a way, the researchers and deep fake creators are unintentionally collaborating.</strong></p>\n\n<p>Please share your opinion on this! I know this may not make any sense, so please correct me if I am wrong :)</p>\n\n<p>(Credit for the idea: <a href=\"/pranavpulijala\">@pranavpulijala</a>)</p>",
      "rawMarkdown": "Hello everyone! I wanted to share a thought with you all. It may be flawed, so take it with a grain of salt.\n\nDFDC challenges participants to build an accurate model to identify deep fake videos. So, in GAN terminology, we are building a \"discriminator\". Based on my understanding, GANs (which are used to generate fake videos) use two networks, a \"discriminator\" and a \"generator\" that compete in a zero-sum game. If we build and open-source better and better \"discriminator\" models, deep fake creators can leverage these models to build better and better \"generators\". Wouldn't that defeat the purpose of the competition by catalyzing the growth of powerful generative models?\n\nThis might result in a chain reaction where researchers keep open-sourcing better \"discriminator\" models and deep fake creators use these models to build better \"generative\" models. **So, in a way, the researchers and deep fake creators are unintentionally collaborating.**\n\nPlease share your opinion on this! I know this may not make any sense, so please correct me if I am wrong :)\n\n(Credit for the idea: @pranavpulijala)",
      "votes": null
    },
    {
      "id": "713271",
      "postDate": "01/08/2020 05:08:20",
      "content": "<p>I just noticed that there is already a <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121206\">post</a> that asks a similar question. So I might delete this post after a few replies come in.</p>",
      "rawMarkdown": "I just noticed that there is already a [post](https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121206) that asks a similar question. So I might delete this post after a few replies come in.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 713271,
      "author_name": "tarunpaparaju",
      "author_url": "",
      "post_date": "01/08/2020 05:08:20",
      "content": "<p>I just noticed that there is already a <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121206\">post</a> that asks a similar question. So I might delete this post after a few replies come in.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "713268": "Hello everyone! I wanted to share a thought with you all. It may be flawed, so take it with a grain of salt.\n\nDFDC challenges participants to build an accurate model to identify deep fake videos. So, in GAN terminology, we are building a \"discriminator\". Based on my understanding, GANs (which are used to generate fake videos) use two networks, a \"discriminator\" and a \"generator\" that compete in a zero-sum game. If we build and open-source better and better \"discriminator\" models, deep fake creators can leverage these models to build better and better \"generators\". Wouldn't that defeat the purpose of the competition by catalyzing the growth of powerful generative models?\n\nThis might result in a chain reaction where researchers keep open-sourcing better \"discriminator\" models and deep fake creators use these models to build better \"generative\" models. **So, in a way, the researchers and deep fake creators are unintentionally collaborating.**\n\nPlease share your opinion on this! I know this may not make any sense, so please correct me if I am wrong :)\n\n(Credit for the idea: @pranavpulijala)",
    "713271": "I just noticed that there is already a [post](https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121206) that asks a similar question. So I might delete this post after a few replies come in."
  },
  "source": "meta"
}