{
  "id": 129095,
  "title": "Questions from a beginner",
  "url": "/competitions/deepfake-detection-challenge/discussion/129095",
  "author_name": "",
  "post_date": "2020-02-05T12:39:14.184542700Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Heya, just started with this competition and I had some questions for which I couldn't find any answers:</p>\n\n<ol>\n<li>Are solutions that don't use machine learning allowed?</li>\n<li>How many videos are expected to be in the private test set?</li>\n<li>Are private test set videos similar to those in the sets we've been supplied, or will they be actual videos found in the wild? </li>\n<li>Are there any videos where just the sound is changed? Can we expect videos like these to be in the private test set and considered deepfakes?</li>\n<li>Are all videos we need to test guaranteed to be .mp4 files?</li>\n<li>I'm trying to get the 'hello world' of predictions in (0.5 on everything), but am getting \"Submission Scoring Error\" even though my generated submission.csv looks the same as the sample submission file. With only 2 submissions a day testing blindly is not great, so could anyone have a look and see what I'm doing wrong? Added both my notebook code and the submission it generates as attachments</li>\n</ol>",
  "messages": [
    {
      "id": "737500",
      "postDate": "02/05/2020 12:39:14",
      "content": "<p>Heya, just started with this competition and I had some questions for which I couldn't find any answers:</p>\n\n<ol>\n<li>Are solutions that don't use machine learning allowed?</li>\n<li>How many videos are expected to be in the private test set?</li>\n<li>Are private test set videos similar to those in the sets we've been supplied, or will they be actual videos found in the wild? </li>\n<li>Are there any videos where just the sound is changed? Can we expect videos like these to be in the private test set and considered deepfakes?</li>\n<li>Are all videos we need to test guaranteed to be .mp4 files?</li>\n<li>I'm trying to get the 'hello world' of predictions in (0.5 on everything), but am getting \"Submission Scoring Error\" even though my generated submission.csv looks the same as the sample submission file. With only 2 submissions a day testing blindly is not great, so could anyone have a look and see what I'm doing wrong? Added both my notebook code and the submission it generates as attachments</li>\n</ol>",
      "rawMarkdown": "Heya, just started with this competition and I had some questions for which I couldn't find any answers:\n\n1. Are solutions that don't use machine learning allowed?\n2. How many videos are expected to be in the private test set?\n3. Are private test set videos similar to those in the sets we've been supplied, or will they be actual videos found in the wild? \n4. Are there any videos where just the sound is changed? Can we expect videos like these to be in the private test set and considered deepfakes?\n5. Are all videos we need to test guaranteed to be .mp4 files?\n6. I'm trying to get the 'hello world' of predictions in (0.5 on everything), but am getting \"Submission Scoring Error\" even though my generated submission.csv looks the same as the sample submission file. With only 2 submissions a day testing blindly is not great, so could anyone have a look and see what I'm doing wrong? Added both my notebook code and the submission it generates as attachments",
      "votes": null
    },
    {
      "id": "737574",
      "postDate": "02/05/2020 14:50:37",
      "content": "<p>1) Yep\n2) The public test set which is used to calculate the leaderboard at the moment is ~4000 videos. THe private test set, used eventually to decide the prizes, we don't know.\n3) We don't know.\n4) I don't actually know the answer to this, but I do know that only 10% of 'fake' videos have audio which is even slightly different when analysed as a spectrogram.\n5) All the training set, public validation set, and public test set videos are MP4 files, so this is a reasonable assumption.\n6) Your submission.csv file has 800 videos in it - it should be 400 videos. Are you predicting the training files?</p>\n\n<p>The following videos beginning with 'a' are in my submissions.csv file:\naassnaulhq.mp4\naayfryxljh.mp4\nacazlolrpz.mp4\nadohdulfwb.mp4\nahjnxtiamx.mp4\najiyrjfyzp.mp4\naktnlyqpah.mp4\nalrtntfxtd.mp4\naomqqjipcp.mp4\napedduehoy.mp4\napvzjkvnwn.mp4\naqrsylrzgi.mp4\naxfhbpkdlc.mp4\nayipraspbn.mp4</p>\n\n<p>Your submission however has many more:</p>\n\n<p>aagfhgtpmv.mp4\naapnvogymq.mp4\naassnaulhq.mp4\naayfryxljh.mp4\nabarnvbtwb.mp4\nabofeumbvv.mp4\nabqwwspghj.mp4\nacazlolrpz.mp4\nacifjvzvpm.mp4\nacqfdwsrhi.mp4\nacxnxvbsxk.mp4\nacxwigylke.mp4\naczrgyricp.mp4\nadhsbajydo.mp4\nadohdulfwb.mp4\nadohikbdaz.mp4\nadylbeequz.mp4\naelfnikyqj.mp4\naelzhcnwgf.mp4\naettqgevhz.mp4\naevrfsexku.mp4\nafoovlsmtx.mp4\nagdkmztvby.mp4\nagqphdxmwt.mp4\nagrmhtjdlk.mp4\nahbweevwpv.mp4\nahdbuwqxit.mp4\nahfazfbntc.mp4\nahjnxtiamx.mp4\nahqqqilsxt.mp4\naipfdnwpoo.mp4\najiyrjfyzp.mp4\najqslcypsw.mp4\najwpjhrbcv.mp4\naklqzsddfl.mp4\naknbdpmgua.mp4\naknmpoonls.mp4\naktnlyqpah.mp4\nakvmwkdyuv.mp4\nakxoopqjqz.mp4\nakzbnazxtz.mp4\naladcziidp.mp4\nalaijyygdv.mp4\nalninxcyhg.mp4\nalrtntfxtd.mp4\naltziddtxi.mp4\nalvgwypubw.mp4\namaivqofda.mp4\namowujxmzc.mp4\nandaxzscny.mp4\naneclqfpbt.mp4\nanpuvshzoo.mp4\naomqqjipcp.mp4\naorjvbyxhw.mp4\napatcsqejh.mp4\napedduehoy.mp4\napgjqzkoma.mp4\napogckdfrz.mp4\napvzjkvnwn.mp4\naqpnvjhuzw.mp4\naqrsylrzgi.mp4\narkroixhey.mp4\narlmiizoob.mp4\narrhsnjqku.mp4\nasaxgevnnp.mp4\nasdpeebotb.mp4\naslsvlvpth.mp4\nasmpfjfzif.mp4\nasvcrfdpnq.mp4\natkdltyyen.mp4\natvmxvwyns.mp4\natxvxouljq.mp4\natyntldecu.mp4\natzdznmder.mp4\naufmsmnoye.mp4\naugtsuxpzc.mp4\navfitoutyn.mp4\navgiuextiz.mp4\navibnnhwhp.mp4\navmjormvsx.mp4\navnqydkqjj.mp4\navssvvsdhz.mp4\navtycwsgyb.mp4\navvdgsennp.mp4\navywawptfc.mp4\nawhmfnnjih.mp4\nawnwkrqibf.mp4\nawukslzjra.mp4\naxczxisdtb.mp4\naxfhbpkdlc.mp4\naxntxmycwd.mp4\naxoygtekut.mp4\naxwgcsyphv.mp4\naxwovszumc.mp4\naybgughjxh.mp4\naybumesmpk.mp4\nayipraspbn.mp4\nayqvfdhslr.mp4\naytzyidmgs.mp4\nazpuxunqyo.mp4\nazsmewqghg.mp4</p>",
      "rawMarkdown": "1) Yep\n2) The public test set which is used to calculate the leaderboard at the moment is ~4000 videos. THe private test set, used eventually to decide the prizes, we don't know.\n3) We don't know.\n4) I don't actually know the answer to this, but I do know that only 10% of 'fake' videos have audio which is even slightly different when analysed as a spectrogram.\n5) All the training set, public validation set, and public test set videos are MP4 files, so this is a reasonable assumption.\n6) Your submission.csv file has 800 videos in it - it should be 400 videos. Are you predicting the training files?\n\nThe following videos beginning with 'a' are in my submissions.csv file:\naassnaulhq.mp4\naayfryxljh.mp4\nacazlolrpz.mp4\nadohdulfwb.mp4\nahjnxtiamx.mp4\najiyrjfyzp.mp4\naktnlyqpah.mp4\nalrtntfxtd.mp4\naomqqjipcp.mp4\napedduehoy.mp4\napvzjkvnwn.mp4\naqrsylrzgi.mp4\naxfhbpkdlc.mp4\nayipraspbn.mp4\n\nYour submission however has many more:\n\naagfhgtpmv.mp4\naapnvogymq.mp4\naassnaulhq.mp4\naayfryxljh.mp4\nabarnvbtwb.mp4\nabofeumbvv.mp4\nabqwwspghj.mp4\nacazlolrpz.mp4\nacifjvzvpm.mp4\nacqfdwsrhi.mp4\nacxnxvbsxk.mp4\nacxwigylke.mp4\naczrgyricp.mp4\nadhsbajydo.mp4\nadohdulfwb.mp4\nadohikbdaz.mp4\nadylbeequz.mp4\naelfnikyqj.mp4\naelzhcnwgf.mp4\naettqgevhz.mp4\naevrfsexku.mp4\nafoovlsmtx.mp4\nagdkmztvby.mp4\nagqphdxmwt.mp4\nagrmhtjdlk.mp4\nahbweevwpv.mp4\nahdbuwqxit.mp4\nahfazfbntc.mp4\nahjnxtiamx.mp4\nahqqqilsxt.mp4\naipfdnwpoo.mp4\najiyrjfyzp.mp4\najqslcypsw.mp4\najwpjhrbcv.mp4\naklqzsddfl.mp4\naknbdpmgua.mp4\naknmpoonls.mp4\naktnlyqpah.mp4\nakvmwkdyuv.mp4\nakxoopqjqz.mp4\nakzbnazxtz.mp4\naladcziidp.mp4\nalaijyygdv.mp4\nalninxcyhg.mp4\nalrtntfxtd.mp4\naltziddtxi.mp4\nalvgwypubw.mp4\namaivqofda.mp4\namowujxmzc.mp4\nandaxzscny.mp4\naneclqfpbt.mp4\nanpuvshzoo.mp4\naomqqjipcp.mp4\naorjvbyxhw.mp4\napatcsqejh.mp4\napedduehoy.mp4\napgjqzkoma.mp4\napogckdfrz.mp4\napvzjkvnwn.mp4\naqpnvjhuzw.mp4\naqrsylrzgi.mp4\narkroixhey.mp4\narlmiizoob.mp4\narrhsnjqku.mp4\nasaxgevnnp.mp4\nasdpeebotb.mp4\naslsvlvpth.mp4\nasmpfjfzif.mp4\nasvcrfdpnq.mp4\natkdltyyen.mp4\natvmxvwyns.mp4\natxvxouljq.mp4\natyntldecu.mp4\natzdznmder.mp4\naufmsmnoye.mp4\naugtsuxpzc.mp4\navfitoutyn.mp4\navgiuextiz.mp4\navibnnhwhp.mp4\navmjormvsx.mp4\navnqydkqjj.mp4\navssvvsdhz.mp4\navtycwsgyb.mp4\navvdgsennp.mp4\navywawptfc.mp4\nawhmfnnjih.mp4\nawnwkrqibf.mp4\nawukslzjra.mp4\naxczxisdtb.mp4\naxfhbpkdlc.mp4\naxntxmycwd.mp4\naxoygtekut.mp4\naxwgcsyphv.mp4\naxwovszumc.mp4\naybgughjxh.mp4\naybumesmpk.mp4\nayipraspbn.mp4\nayqvfdhslr.mp4\naytzyidmgs.mp4\nazpuxunqyo.mp4\nazsmewqghg.mp4",
      "votes": null
    },
    {
      "id": "737654",
      "postDate": "02/05/2020 16:13:29",
      "content": "<p>Thanks a ton! That must be the issue. I'm currently grabbing every file that's under /input. I suppose I just have to walk through /kaggle/input/deepfake-detection-challenge/test_videos  then?  Can I assume that that's also the folder I need to target for the private test data?</p>\n\n<p>Shame it's not known how many videos there will be in the private set, that feels like it'd be important information to determine whether your program is fast enough</p>",
      "rawMarkdown": "Thanks a ton! That must be the issue. I'm currently grabbing every file that's under /input. I suppose I just have to walk through /kaggle/input/deepfake-detection-challenge/test_videos  then?  Can I assume that that's also the folder I need to target for the private test data?\n\nShame it's not known how many videos there will be in the private set, that feels like it'd be important information to determine whether your program is fast enough",
      "votes": null
    },
    {
      "id": "737660",
      "postDate": "02/05/2020 16:18:27",
      "content": "<p>Yes that's right you should just be using test_videos.</p>\n\n<p>Your code needs to be fast enough to process the 4000 public test set videos within 9 hours for the leaderboard, but if it can do that you don't need to worry about the private test set size as my understanding is the data is held and processed outside of kaggle and therefore any time restrictions?</p>",
      "rawMarkdown": "Yes that's right you should just be using test_videos.\n\nYour code needs to be fast enough to process the 4000 public test set videos within 9 hours for the leaderboard, but if it can do that you don't need to worry about the private test set size as my understanding is the data is held and processed outside of kaggle and therefore any time restrictions?",
      "votes": null
    },
    {
      "id": "737913",
      "postDate": "02/05/2020 22:51:30",
      "content": "<p>Regarding 4, i have tested most of the training set and could not find videos with only modified audio </p>",
      "rawMarkdown": "Regarding 4, i have tested most of the training set and could not find videos with only modified audio",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 737574,
      "author_name": "jamesphoward",
      "author_url": "",
      "post_date": "02/05/2020 14:50:37",
      "content": "<p>1) Yep\n2) The public test set which is used to calculate the leaderboard at the moment is ~4000 videos. THe private test set, used eventually to decide the prizes, we don't know.\n3) We don't know.\n4) I don't actually know the answer to this, but I do know that only 10% of 'fake' videos have audio which is even slightly different when analysed as a spectrogram.\n5) All the training set, public validation set, and public test set videos are MP4 files, so this is a reasonable assumption.\n6) Your submission.csv file has 800 videos in it - it should be 400 videos. Are you predicting the training files?</p>\n\n<p>The following videos beginning with 'a' are in my submissions.csv file:\naassnaulhq.mp4\naayfryxljh.mp4\nacazlolrpz.mp4\nadohdulfwb.mp4\nahjnxtiamx.mp4\najiyrjfyzp.mp4\naktnlyqpah.mp4\nalrtntfxtd.mp4\naomqqjipcp.mp4\napedduehoy.mp4\napvzjkvnwn.mp4\naqrsylrzgi.mp4\naxfhbpkdlc.mp4\nayipraspbn.mp4</p>\n\n<p>Your submission however has many more:</p>\n\n<p>aagfhgtpmv.mp4\naapnvogymq.mp4\naassnaulhq.mp4\naayfryxljh.mp4\nabarnvbtwb.mp4\nabofeumbvv.mp4\nabqwwspghj.mp4\nacazlolrpz.mp4\nacifjvzvpm.mp4\nacqfdwsrhi.mp4\nacxnxvbsxk.mp4\nacxwigylke.mp4\naczrgyricp.mp4\nadhsbajydo.mp4\nadohdulfwb.mp4\nadohikbdaz.mp4\nadylbeequz.mp4\naelfnikyqj.mp4\naelzhcnwgf.mp4\naettqgevhz.mp4\naevrfsexku.mp4\nafoovlsmtx.mp4\nagdkmztvby.mp4\nagqphdxmwt.mp4\nagrmhtjdlk.mp4\nahbweevwpv.mp4\nahdbuwqxit.mp4\nahfazfbntc.mp4\nahjnxtiamx.mp4\nahqqqilsxt.mp4\naipfdnwpoo.mp4\najiyrjfyzp.mp4\najqslcypsw.mp4\najwpjhrbcv.mp4\naklqzsddfl.mp4\naknbdpmgua.mp4\naknmpoonls.mp4\naktnlyqpah.mp4\nakvmwkdyuv.mp4\nakxoopqjqz.mp4\nakzbnazxtz.mp4\naladcziidp.mp4\nalaijyygdv.mp4\nalninxcyhg.mp4\nalrtntfxtd.mp4\naltziddtxi.mp4\nalvgwypubw.mp4\namaivqofda.mp4\namowujxmzc.mp4\nandaxzscny.mp4\naneclqfpbt.mp4\nanpuvshzoo.mp4\naomqqjipcp.mp4\naorjvbyxhw.mp4\napatcsqejh.mp4\napedduehoy.mp4\napgjqzkoma.mp4\napogckdfrz.mp4\napvzjkvnwn.mp4\naqpnvjhuzw.mp4\naqrsylrzgi.mp4\narkroixhey.mp4\narlmiizoob.mp4\narrhsnjqku.mp4\nasaxgevnnp.mp4\nasdpeebotb.mp4\naslsvlvpth.mp4\nasmpfjfzif.mp4\nasvcrfdpnq.mp4\natkdltyyen.mp4\natvmxvwyns.mp4\natxvxouljq.mp4\natyntldecu.mp4\natzdznmder.mp4\naufmsmnoye.mp4\naugtsuxpzc.mp4\navfitoutyn.mp4\navgiuextiz.mp4\navibnnhwhp.mp4\navmjormvsx.mp4\navnqydkqjj.mp4\navssvvsdhz.mp4\navtycwsgyb.mp4\navvdgsennp.mp4\navywawptfc.mp4\nawhmfnnjih.mp4\nawnwkrqibf.mp4\nawukslzjra.mp4\naxczxisdtb.mp4\naxfhbpkdlc.mp4\naxntxmycwd.mp4\naxoygtekut.mp4\naxwgcsyphv.mp4\naxwovszumc.mp4\naybgughjxh.mp4\naybumesmpk.mp4\nayipraspbn.mp4\nayqvfdhslr.mp4\naytzyidmgs.mp4\nazpuxunqyo.mp4\nazsmewqghg.mp4</p>",
      "votes": null,
      "replies": [
        {
          "id": 737654,
          "author_name": "danmctree",
          "author_url": "",
          "post_date": "02/05/2020 16:13:29",
          "content": "<p>Thanks a ton! That must be the issue. I'm currently grabbing every file that's under /input. I suppose I just have to walk through /kaggle/input/deepfake-detection-challenge/test_videos  then?  Can I assume that that's also the folder I need to target for the private test data?</p>\n\n<p>Shame it's not known how many videos there will be in the private set, that feels like it'd be important information to determine whether your program is fast enough</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 737660,
          "author_name": "jamesphoward",
          "author_url": "",
          "post_date": "02/05/2020 16:18:27",
          "content": "<p>Yes that's right you should just be using test_videos.</p>\n\n<p>Your code needs to be fast enough to process the 4000 public test set videos within 9 hours for the leaderboard, but if it can do that you don't need to worry about the private test set size as my understanding is the data is held and processed outside of kaggle and therefore any time restrictions?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 737913,
          "author_name": "moshel",
          "author_url": "",
          "post_date": "02/05/2020 22:51:30",
          "content": "<p>Regarding 4, i have tested most of the training set and could not find videos with only modified audio </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "737500": "Heya, just started with this competition and I had some questions for which I couldn't find any answers:\n\n1. Are solutions that don't use machine learning allowed?\n2. How many videos are expected to be in the private test set?\n3. Are private test set videos similar to those in the sets we've been supplied, or will they be actual videos found in the wild? \n4. Are there any videos where just the sound is changed? Can we expect videos like these to be in the private test set and considered deepfakes?\n5. Are all videos we need to test guaranteed to be .mp4 files?\n6. I'm trying to get the 'hello world' of predictions in (0.5 on everything), but am getting \"Submission Scoring Error\" even though my generated submission.csv looks the same as the sample submission file. With only 2 submissions a day testing blindly is not great, so could anyone have a look and see what I'm doing wrong? Added both my notebook code and the submission it generates as attachments",
    "737574": "1) Yep\n2) The public test set which is used to calculate the leaderboard at the moment is ~4000 videos. THe private test set, used eventually to decide the prizes, we don't know.\n3) We don't know.\n4) I don't actually know the answer to this, but I do know that only 10% of 'fake' videos have audio which is even slightly different when analysed as a spectrogram.\n5) All the training set, public validation set, and public test set videos are MP4 files, so this is a reasonable assumption.\n6) Your submission.csv file has 800 videos in it - it should be 400 videos. Are you predicting the training files?\n\nThe following videos beginning with 'a' are in my submissions.csv file:\naassnaulhq.mp4\naayfryxljh.mp4\nacazlolrpz.mp4\nadohdulfwb.mp4\nahjnxtiamx.mp4\najiyrjfyzp.mp4\naktnlyqpah.mp4\nalrtntfxtd.mp4\naomqqjipcp.mp4\napedduehoy.mp4\napvzjkvnwn.mp4\naqrsylrzgi.mp4\naxfhbpkdlc.mp4\nayipraspbn.mp4\n\nYour submission however has many more:\n\naagfhgtpmv.mp4\naapnvogymq.mp4\naassnaulhq.mp4\naayfryxljh.mp4\nabarnvbtwb.mp4\nabofeumbvv.mp4\nabqwwspghj.mp4\nacazlolrpz.mp4\nacifjvzvpm.mp4\nacqfdwsrhi.mp4\nacxnxvbsxk.mp4\nacxwigylke.mp4\naczrgyricp.mp4\nadhsbajydo.mp4\nadohdulfwb.mp4\nadohikbdaz.mp4\nadylbeequz.mp4\naelfnikyqj.mp4\naelzhcnwgf.mp4\naettqgevhz.mp4\naevrfsexku.mp4\nafoovlsmtx.mp4\nagdkmztvby.mp4\nagqphdxmwt.mp4\nagrmhtjdlk.mp4\nahbweevwpv.mp4\nahdbuwqxit.mp4\nahfazfbntc.mp4\nahjnxtiamx.mp4\nahqqqilsxt.mp4\naipfdnwpoo.mp4\najiyrjfyzp.mp4\najqslcypsw.mp4\najwpjhrbcv.mp4\naklqzsddfl.mp4\naknbdpmgua.mp4\naknmpoonls.mp4\naktnlyqpah.mp4\nakvmwkdyuv.mp4\nakxoopqjqz.mp4\nakzbnazxtz.mp4\naladcziidp.mp4\nalaijyygdv.mp4\nalninxcyhg.mp4\nalrtntfxtd.mp4\naltziddtxi.mp4\nalvgwypubw.mp4\namaivqofda.mp4\namowujxmzc.mp4\nandaxzscny.mp4\naneclqfpbt.mp4\nanpuvshzoo.mp4\naomqqjipcp.mp4\naorjvbyxhw.mp4\napatcsqejh.mp4\napedduehoy.mp4\napgjqzkoma.mp4\napogckdfrz.mp4\napvzjkvnwn.mp4\naqpnvjhuzw.mp4\naqrsylrzgi.mp4\narkroixhey.mp4\narlmiizoob.mp4\narrhsnjqku.mp4\nasaxgevnnp.mp4\nasdpeebotb.mp4\naslsvlvpth.mp4\nasmpfjfzif.mp4\nasvcrfdpnq.mp4\natkdltyyen.mp4\natvmxvwyns.mp4\natxvxouljq.mp4\natyntldecu.mp4\natzdznmder.mp4\naufmsmnoye.mp4\naugtsuxpzc.mp4\navfitoutyn.mp4\navgiuextiz.mp4\navibnnhwhp.mp4\navmjormvsx.mp4\navnqydkqjj.mp4\navssvvsdhz.mp4\navtycwsgyb.mp4\navvdgsennp.mp4\navywawptfc.mp4\nawhmfnnjih.mp4\nawnwkrqibf.mp4\nawukslzjra.mp4\naxczxisdtb.mp4\naxfhbpkdlc.mp4\naxntxmycwd.mp4\naxoygtekut.mp4\naxwgcsyphv.mp4\naxwovszumc.mp4\naybgughjxh.mp4\naybumesmpk.mp4\nayipraspbn.mp4\nayqvfdhslr.mp4\naytzyidmgs.mp4\nazpuxunqyo.mp4\nazsmewqghg.mp4",
    "737654": "Thanks a ton! That must be the issue. I'm currently grabbing every file that's under /input. I suppose I just have to walk through /kaggle/input/deepfake-detection-challenge/test_videos  then?  Can I assume that that's also the folder I need to target for the private test data?\n\nShame it's not known how many videos there will be in the private set, that feels like it'd be important information to determine whether your program is fast enough",
    "737660": "Yes that's right you should just be using test_videos.\n\nYour code needs to be fast enough to process the 4000 public test set videos within 9 hours for the leaderboard, but if it can do that you don't need to worry about the private test set size as my understanding is the data is held and processed outside of kaggle and therefore any time restrictions?",
    "737913": "Regarding 4, i have tested most of the training set and could not find videos with only modified audio"
  },
  "source": "meta"
}