{
  "id": 39897,
  "title": "Data visualization",
  "url": "/competitions/carvana-image-masking-challenge/discussion/39897",
  "author_name": "Artem.Sanakoev",
  "post_date": "2017-09-23T11:44:53.597000",
  "votes": 13,
  "comment_count": 16,
  "views": 0,
  "content": "<p>I have visualized predicts of my network (lb 9971) as gif. <br>\nIf you have cool visualizations to show, please share :)</p>",
  "messages": [
    {
      "id": 223759,
      "postDate": "2017-09-23T11:44:53.597Z",
      "content": "<p>I have visualized predicts of my network (lb 9971) as gif. <br>\nIf you have cool visualizations to show, please share :)</p>",
      "rawMarkdown": "I have visualized predicts of my network (lb 9971) as gif.   \nIf you have cool visualizations to show, please share :)\n",
      "votes": 13
    },
    {
      "id": 224901,
      "postDate": "2017-09-27T22:11:51.030Z",
      "content": "<p>I made a video based on my solution:</p>\n\n<p><a href=\"https://www.youtube.com/watch?v=hwCUY4mwX1I\">https://www.youtube.com/watch?v=hwCUY4mwX1I</a></p>\n\n<p>P.S. Subscribe to add motivation for new DL videos ))</p>",
      "rawMarkdown": "I made a video based on my solution:\n\nhttps://www.youtube.com/watch?v=hwCUY4mwX1I\n\nP.S. Subscribe to add motivation for new DL videos ))\n",
      "votes": 11
    },
    {
      "id": 223789,
      "postDate": "2017-09-23T15:42:48.560Z",
      "content": "<p>and here (contour mode)</p>",
      "rawMarkdown": "and here (contour mode)",
      "votes": 1
    },
    {
      "id": 224787,
      "postDate": "2017-09-27T17:00:50.920Z",
      "content": "<p>looking at ensemble results:</p>",
      "rawMarkdown": "looking at ensemble results:"
    },
    {
      "id": 224705,
      "postDate": "2017-09-27T11:05:37.180Z",
      "content": "<p>visualization with raw probability score and threshold mask (another version)</p>",
      "rawMarkdown": "visualization with raw probability score and threshold mask (another version)"
    },
    {
      "id": 224704,
      "postDate": "2017-09-27T11:02:47.350Z",
      "content": "<p>visualization with raw probability score and threshold mask </p>",
      "rawMarkdown": "visualization with raw probability score and threshold mask "
    },
    {
      "id": 224667,
      "postDate": "2017-09-27T07:07:59.090Z",
      "content": "<p>Nice visualization, congrats!</p>",
      "rawMarkdown": "Nice visualization, congrats!"
    },
    {
      "id": 224000,
      "postDate": "2017-09-24T17:53:23.750Z",
      "content": "<p>Hi guys, I have a questions. Regarding the holes tin the spin wheels, isn't that considered as an error? For instance I know that in the training masks these holes exist on wheels and other parts but eventually don't we want the predicted mask to be whole, i.e. without discontinuities?</p>",
      "rawMarkdown": "Hi guys, I have a questions. Regarding the holes tin the spin wheels, isn't that considered as an error? For instance I know that in the training masks these holes exist on wheels and other parts but eventually don't we want the predicted mask to be whole, i.e. without discontinuities?",
      "replies": [
        {
          "id": 224101,
          "postDate": "2017-09-25T05:50:33.277Z",
          "content": "<p>I don't think so since there is no necessity for it to continuous.The motto of masking is to get the required object, separating it from the background. So, if the wheels have holes where background is visible so it needs to be eliminated from the mask.</p>",
          "rawMarkdown": "I don't think so since there is no necessity for it to continuous.The motto of masking is to get the required object, separating it from the background. So, if the wheels have holes where background is visible so it needs to be eliminated from the mask."
        }
      ]
    },
    {
      "id": 223933,
      "postDate": "2017-09-24T10:53:04.777Z",
      "content": "<p>great work! </p>",
      "rawMarkdown": "great work! "
    },
    {
      "id": 223899,
      "postDate": "2017-09-24T05:58:28.247Z",
      "content": "<p>Is it the result of a single network? I.e your single network has lb 0.9971?</p>",
      "rawMarkdown": "Is it the result of a single network? I.e your single network has lb 0.9971?",
      "replies": [
        {
          "id": 223951,
          "postDate": "2017-09-24T12:18:40.137Z",
          "content": "<p>These are results of ensemle of my networks, ensemble gives 0.9971 on lb.</p>",
          "rawMarkdown": "These are results of ensemle of my networks, ensemble gives 0.9971 on lb."
        },
        {
          "id": 224103,
          "postDate": "2017-09-25T06:05:27.850Z",
          "content": "<p>The difference in @Artem and @Heng is the size of the image displayed. </p>\n\n<p>@Artem Are you not using resizing i.e reducing the size when training/predicting?</p>",
          "rawMarkdown": "The difference in @Artem and @Heng is the size of the image displayed. \n\n@Artem Are you not using resizing i.e reducing the size when training/predicting?"
        }
      ]
    },
    {
      "id": 223851,
      "postDate": "2017-09-23T21:36:26.193Z",
      "content": "<p>The predictions are really nice! </p>",
      "rawMarkdown": "The predictions are really nice! "
    },
    {
      "id": 223788,
      "postDate": "2017-09-23T15:41:48.447Z",
      "content": "<p>here is my version (mask mode)</p>",
      "rawMarkdown": "here is my version (mask mode)"
    },
    {
      "id": 223778,
      "postDate": "2017-09-23T14:26:57.753Z",
      "content": "<p>Your predicted masks are really nice and precise. </p>",
      "rawMarkdown": "Your predicted masks are really nice and precise. "
    },
    {
      "id": 224297,
      "postDate": "2017-09-25T20:43:20.753Z",
      "content": "<p>solid and neat!</p>",
      "rawMarkdown": "solid and neat!",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 224901,
      "author_name": "ZFTurbo",
      "author_url": "",
      "post_date": "2017-09-27T22:11:51.030000",
      "content": "<p>I made a video based on my solution:</p>\n\n<p><a href=\"https://www.youtube.com/watch?v=hwCUY4mwX1I\">https://www.youtube.com/watch?v=hwCUY4mwX1I</a></p>\n\n<p>P.S. Subscribe to add motivation for new DL videos ))</p>",
      "votes": 11,
      "replies": []
    },
    {
      "id": 223789,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-23T15:42:48.560000",
      "content": "<p>and here (contour mode)</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 224787,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-27T17:00:50.920000",
      "content": "<p>looking at ensemble results:</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 224705,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-27T11:05:37.180000",
      "content": "<p>visualization with raw probability score and threshold mask (another version)</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 224704,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-27T11:02:47.350000",
      "content": "<p>visualization with raw probability score and threshold mask </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 224667,
      "author_name": "",
      "author_url": "",
      "post_date": "2017-09-27T07:07:59.090000",
      "content": "<p>Nice visualization, congrats!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 224000,
      "author_name": "kirk",
      "author_url": "",
      "post_date": "2017-09-24T17:53:23.750000",
      "content": "<p>Hi guys, I have a questions. Regarding the holes tin the spin wheels, isn't that considered as an error? For instance I know that in the training masks these holes exist on wheels and other parts but eventually don't we want the predicted mask to be whole, i.e. without discontinuities?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 224101,
          "author_name": "remidi",
          "author_url": "",
          "post_date": "2017-09-25T05:50:33.277000",
          "content": "<p>I don't think so since there is no necessity for it to continuous.The motto of masking is to get the required object, separating it from the background. So, if the wheels have holes where background is visible so it needs to be eliminated from the mask.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 223933,
      "author_name": "IJKos",
      "author_url": "",
      "post_date": "2017-09-24T10:53:04.777000",
      "content": "<p>great work! </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 223899,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-24T05:58:28.247000",
      "content": "<p>Is it the result of a single network? I.e your single network has lb 0.9971?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 223951,
          "author_name": "Artem.Sanakoev",
          "author_url": "",
          "post_date": "2017-09-24T12:18:40.137000",
          "content": "<p>These are results of ensemle of my networks, ensemble gives 0.9971 on lb.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 224103,
          "author_name": "remidi",
          "author_url": "",
          "post_date": "2017-09-25T06:05:27.850000",
          "content": "<p>The difference in @Artem and @Heng is the size of the image displayed. </p>\n\n<p>@Artem Are you not using resizing i.e reducing the size when training/predicting?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 223851,
      "author_name": "Junhong Xu",
      "author_url": "",
      "post_date": "2017-09-23T21:36:26.193000",
      "content": "<p>The predictions are really nice! </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 223788,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2017-09-23T15:41:48.447000",
      "content": "<p>here is my version (mask mode)</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 223778,
      "author_name": "Aramis",
      "author_url": "",
      "post_date": "2017-09-23T14:26:57.753000",
      "content": "<p>Your predicted masks are really nice and precise. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 224297,
      "author_name": "",
      "author_url": "",
      "post_date": "2017-09-25T20:43:20.753000",
      "content": "<p>solid and neat!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "223759": "I have visualized predicts of my network (lb 9971) as gif.   \nIf you have cool visualizations to show, please share :)\n",
    "224901": "I made a video based on my solution:\n\nhttps://www.youtube.com/watch?v=hwCUY4mwX1I\n\nP.S. Subscribe to add motivation for new DL videos ))\n",
    "223789": "and here (contour mode)",
    "224787": "looking at ensemble results:",
    "224705": "visualization with raw probability score and threshold mask (another version)",
    "224704": "visualization with raw probability score and threshold mask ",
    "224667": "Nice visualization, congrats!",
    "224000": "Hi guys, I have a questions. Regarding the holes tin the spin wheels, isn't that considered as an error? For instance I know that in the training masks these holes exist on wheels and other parts but eventually don't we want the predicted mask to be whole, i.e. without discontinuities?",
    "223933": "great work! ",
    "223899": "Is it the result of a single network? I.e your single network has lb 0.9971?",
    "223851": "The predictions are really nice! ",
    "223788": "here is my version (mask mode)",
    "223778": "Your predicted masks are really nice and precise. ",
    "224297": "solid and neat!"
  }
}