{
  "id": 124713,
  "title": "Private Test Set: Format and length of the videos ?",
  "url": "/competitions/deepfake-detection-challenge/discussion/124713",
  "author_name": "",
  "post_date": "2020-01-06T03:49:43.881212800Z",
  "votes": 13,
  "comment_count": 6,
  "views": 0,
  "content": "<p>In <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started\">Getting Started</a>, Private Test Set is described as \"similar\" in format and nature as the training and public sets.</p>\n\n<p>Can we assume that the format of the video clips in the private test set is .mp4 and the length of those video clips is ~10s with similar fps ?</p>\n\n<p>@philculliton </p>",
  "messages": [
    {
      "id": "711412",
      "postDate": "01/06/2020 03:49:43",
      "content": "<p>In <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started\">Getting Started</a>, Private Test Set is described as \"similar\" in format and nature as the training and public sets.</p>\n\n<p>Can we assume that the format of the video clips in the private test set is .mp4 and the length of those video clips is ~10s with similar fps ?</p>\n\n<p>@philculliton </p>",
      "rawMarkdown": "In [Getting Started](https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started), Private Test Set is described as \"similar\" in format and nature as the training and public sets.\n\nCan we assume that the format of the video clips in the private test set is .mp4 and the length of those video clips is ~10s with similar fps ?\n\n@philculliton",
      "votes": null
    },
    {
      "id": "718030",
      "postDate": "01/13/2020 22:54:08",
      "content": "<p>Yes, the private test will contain clips of ~10s. The private test set will contain videos featuring DeepFakes from the real world (as mentioned in the \"<a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started\">Get started</a>\" section of this competition) therefore we can't guarantee the frame-rate. As mentioned in the <a href=\"https://arxiv.org/abs/1910.08854\">white paper</a> for the preview DFDC dataset, we'd encourage participant to think on adversarial ways on how data may be modified so to train their algorithms to be effective when dealing with organic data (among those adversarial perturbations, you should consider frame-rate modifications such as sub-sampling).</p>",
      "rawMarkdown": "Yes, the private test will contain clips of ~10s. The private test set will contain videos featuring DeepFakes from the real world (as mentioned in the \"[Get started](https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started)\" section of this competition) therefore we can't guarantee the frame-rate. As mentioned in the [white paper](https://arxiv.org/abs/1910.08854) for the preview DFDC dataset, we'd encourage participant to think on adversarial ways on how data may be modified so to train their algorithms to be effective when dealing with organic data (among those adversarial perturbations, you should consider frame-rate modifications such as sub-sampling).",
      "votes": null
    },
    {
      "id": "718036",
      "postDate": "01/13/2020 23:13:29",
      "content": "<p>Thanks for your clarification <a href=\"/cristiancanton\">@cristiancanton</a>. If you may, could you also add an interpretation to \"organic\" or \"real world\" data? Is it possible that the Private Test Set will not contain even a single video of actors similar to the data given in the Training Set?</p>",
      "rawMarkdown": "Thanks for your clarification @cristiancanton. If you may, could you also add an interpretation to \"organic\" or \"real world\" data? Is it possible that the Private Test Set will not contain even a single video of actors similar to the data given in the Training Set?",
      "votes": null
    },
    {
      "id": "718091",
      "postDate": "01/14/2020 01:47:53",
      "content": "<p>thank you Mozaic for the clarification</p>",
      "rawMarkdown": "thank you Mozaic for the clarification",
      "votes": null
    },
    {
      "id": "718125",
      "postDate": "01/14/2020 03:24:46",
      "content": "<p>The Private Test set will contain a mixture of data of many sorts. Among them, DeepFakes like the ones you may encounter on the internet (organic traffic) and data with a similar distribution than the data in the Public Test set. Hope this clarifies this.</p>",
      "rawMarkdown": "The Private Test set will contain a mixture of data of many sorts. Among them, DeepFakes like the ones you may encounter on the internet (organic traffic) and data with a similar distribution than the data in the Public Test set. Hope this clarifies this.",
      "votes": null
    },
    {
      "id": "764355",
      "postDate": "03/05/2020 11:37:16",
      "content": "<p>Yes</p>",
      "rawMarkdown": "Yes",
      "votes": null
    },
    {
      "id": "780217",
      "postDate": "03/20/2020 04:23:53",
      "content": "<p>Will all the videos in the Private Test set be H264 encoded like the ones in the train/test sets provided?</p>",
      "rawMarkdown": "Will all the videos in the Private Test set be H264 encoded like the ones in the train/test sets provided?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 718030,
      "author_name": "cristiancanton",
      "author_url": "",
      "post_date": "01/13/2020 22:54:08",
      "content": "<p>Yes, the private test will contain clips of ~10s. The private test set will contain videos featuring DeepFakes from the real world (as mentioned in the \"<a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started\">Get started</a>\" section of this competition) therefore we can't guarantee the frame-rate. As mentioned in the <a href=\"https://arxiv.org/abs/1910.08854\">white paper</a> for the preview DFDC dataset, we'd encourage participant to think on adversarial ways on how data may be modified so to train their algorithms to be effective when dealing with organic data (among those adversarial perturbations, you should consider frame-rate modifications such as sub-sampling).</p>",
      "votes": null,
      "replies": [
        {
          "id": 718036,
          "author_name": "rafiko1",
          "author_url": "",
          "post_date": "01/13/2020 23:13:29",
          "content": "<p>Thanks for your clarification <a href=\"/cristiancanton\">@cristiancanton</a>. If you may, could you also add an interpretation to \"organic\" or \"real world\" data? Is it possible that the Private Test Set will not contain even a single video of actors similar to the data given in the Training Set?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 718091,
          "author_name": "ravivadapalli",
          "author_url": "",
          "post_date": "01/14/2020 01:47:53",
          "content": "<p>thank you Mozaic for the clarification</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 718125,
          "author_name": "cristiancanton",
          "author_url": "",
          "post_date": "01/14/2020 03:24:46",
          "content": "<p>The Private Test set will contain a mixture of data of many sorts. Among them, DeepFakes like the ones you may encounter on the internet (organic traffic) and data with a similar distribution than the data in the Public Test set. Hope this clarifies this.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 780217,
          "author_name": "kenkrige",
          "author_url": "",
          "post_date": "03/20/2020 04:23:53",
          "content": "<p>Will all the videos in the Private Test set be H264 encoded like the ones in the train/test sets provided?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 764355,
      "author_name": "machrouh",
      "author_url": "",
      "post_date": "03/05/2020 11:37:16",
      "content": "<p>Yes</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "711412": "In [Getting Started](https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started), Private Test Set is described as \"similar\" in format and nature as the training and public sets.\n\nCan we assume that the format of the video clips in the private test set is .mp4 and the length of those video clips is ~10s with similar fps ?\n\n@philculliton",
    "718030": "Yes, the private test will contain clips of ~10s. The private test set will contain videos featuring DeepFakes from the real world (as mentioned in the \"[Get started](https://www.kaggle.com/c/deepfake-detection-challenge/overview/getting-started)\" section of this competition) therefore we can't guarantee the frame-rate. As mentioned in the [white paper](https://arxiv.org/abs/1910.08854) for the preview DFDC dataset, we'd encourage participant to think on adversarial ways on how data may be modified so to train their algorithms to be effective when dealing with organic data (among those adversarial perturbations, you should consider frame-rate modifications such as sub-sampling).",
    "718036": "Thanks for your clarification @cristiancanton. If you may, could you also add an interpretation to \"organic\" or \"real world\" data? Is it possible that the Private Test Set will not contain even a single video of actors similar to the data given in the Training Set?",
    "718091": "thank you Mozaic for the clarification",
    "718125": "The Private Test set will contain a mixture of data of many sorts. Among them, DeepFakes like the ones you may encounter on the internet (organic traffic) and data with a similar distribution than the data in the Public Test set. Hope this clarifies this.",
    "764355": "Yes",
    "780217": "Will all the videos in the Private Test set be H264 encoded like the ones in the train/test sets provided?"
  },
  "source": "meta"
}