{
  "id": 397768,
  "title": "Question about dice-score",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/397768",
  "author_name": "",
  "post_date": "2023-03-27T07:03:51.619851700Z",
  "votes": 1,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I use the module segmentation_model to calculate the dice-score<br>\nand i get the result which is smaller than zero !!?<br>\nbut from plotting  result of predictions , prediction is quite similar to ground truth …<br>\nI don't have any ideas …<br>\nanybody can give me some advices? </p>",
  "messages": [
    {
      "id": "2198592",
      "postDate": "03/27/2023 07:03:51",
      "content": "<p>I use the module segmentation_model to calculate the dice-score<br>\nand i get the result which is smaller than zero !!?<br>\nbut from plotting  result of predictions , prediction is quite similar to ground truth …<br>\nI don't have any ideas …<br>\nanybody can give me some advices? </p>",
      "rawMarkdown": "I use the module segmentation_model to calculate the dice-score\nand i get the result which is smaller than zero !!?\nbut from plotting  result of predictions , prediction is quite similar to ground truth ...\nI don't have any ideas ...\nanybody can give me some advices?",
      "votes": null
    },
    {
      "id": "2199356",
      "postDate": "03/27/2023 16:56:00",
      "content": "<p>Dice score is never negative, you have an additional sign somewhere. More critically, you misunderstood the problem: We are supposed to predict the ink mask, but you predicted the fragment mask which is given…</p>",
      "rawMarkdown": "Dice score is never negative, you have an additional sign somewhere. More critically, you misunderstood the problem: We are supposed to predict the ink mask, but you predicted the fragment mask which is given...",
      "votes": null
    },
    {
      "id": "2200982",
      "postDate": "03/29/2023 00:19:13",
      "content": "<p>i calculate the dice by using prediction and ground truth directly , so i really don't have any idea why it is negative….</p>\n<p>in my understanding,we have to build up a 2D segmentation model to predect the region of ink mask , is it correct?(or we have to predict the region of ink with words !!?)<br>\nIn above , my aim is trying to calculate the dice to evaluate preformance of trained model , soi use the image in test folder to get the mask…<br>\n(If my understanding is wrong , can you point out ? this is my first time to join kaggle…)</p>\n<p>Thanks a lot!</p>",
      "rawMarkdown": "i calculate the dice by using prediction and ground truth directly , so i really don't have any idea why it is negative....\n\nin my understanding,we have to build up a 2D segmentation model to predect the region of ink mask , is it correct?(or we have to predict the region of ink with words !!?)\nIn above , my aim is trying to calculate the dice to evaluate preformance of trained model , soi use the image in test folder to get the mask...\n(If my understanding is wrong , can you point out ? this is my first time to join kaggle...)\n\nThanks a lot!",
      "votes": null
    },
    {
      "id": "2201176",
      "postDate": "03/29/2023 05:37:23",
      "content": "<p>I would like to make sure that the final predictions looks like the mask.png in test folder, rights? or mask.png in train folders or inklabels.png in train folders ? in 2D mask?</p>",
      "rawMarkdown": "I would like to make sure that the final predictions looks like the mask.png in test folder, rights? or mask.png in train folders or inklabels.png in train folders ? in 2D mask?",
      "votes": null
    },
    {
      "id": "2202019",
      "postDate": "03/29/2023 17:16:55",
      "content": "<p>No, mask.png only shows which pixels contain data. The final predictions should look like inklabels.png.</p>",
      "rawMarkdown": "No, mask.png only shows which pixels contain data. The final predictions should look like inklabels.png.",
      "votes": null
    },
    {
      "id": "2202333",
      "postDate": "03/30/2023 00:05:05",
      "content": "<p>i see ! Thanks for your replying! It is important to me！</p>",
      "rawMarkdown": "i see ! Thanks for your replying! It is important to me！",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2199356,
      "author_name": "junxhuang",
      "author_url": "",
      "post_date": "03/27/2023 16:56:00",
      "content": "<p>Dice score is never negative, you have an additional sign somewhere. More critically, you misunderstood the problem: We are supposed to predict the ink mask, but you predicted the fragment mask which is given…</p>",
      "votes": null,
      "replies": [
        {
          "id": 2200982,
          "author_name": "chushengchen",
          "author_url": "",
          "post_date": "03/29/2023 00:19:13",
          "content": "<p>i calculate the dice by using prediction and ground truth directly , so i really don't have any idea why it is negative….</p>\n<p>in my understanding,we have to build up a 2D segmentation model to predect the region of ink mask , is it correct?(or we have to predict the region of ink with words !!?)<br>\nIn above , my aim is trying to calculate the dice to evaluate preformance of trained model , soi use the image in test folder to get the mask…<br>\n(If my understanding is wrong , can you point out ? this is my first time to join kaggle…)</p>\n<p>Thanks a lot!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2201176,
          "author_name": "chushengchen",
          "author_url": "",
          "post_date": "03/29/2023 05:37:23",
          "content": "<p>I would like to make sure that the final predictions looks like the mask.png in test folder, rights? or mask.png in train folders or inklabels.png in train folders ? in 2D mask?</p>",
          "votes": null,
          "replies": [
            {
              "id": 2202019,
              "author_name": "jpposma",
              "author_url": "",
              "post_date": "03/29/2023 17:16:55",
              "content": "<p>No, mask.png only shows which pixels contain data. The final predictions should look like inklabels.png.</p>",
              "votes": null,
              "replies": [
                {
                  "id": 2202333,
                  "author_name": "chushengchen",
                  "author_url": "",
                  "post_date": "03/30/2023 00:05:05",
                  "content": "<p>i see ! Thanks for your replying! It is important to me！</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2198592": "I use the module segmentation_model to calculate the dice-score\nand i get the result which is smaller than zero !!?\nbut from plotting  result of predictions , prediction is quite similar to ground truth ...\nI don't have any ideas ...\nanybody can give me some advices?",
    "2199356": "Dice score is never negative, you have an additional sign somewhere. More critically, you misunderstood the problem: We are supposed to predict the ink mask, but you predicted the fragment mask which is given...",
    "2200982": "i calculate the dice by using prediction and ground truth directly , so i really don't have any idea why it is negative....\n\nin my understanding,we have to build up a 2D segmentation model to predect the region of ink mask , is it correct?(or we have to predict the region of ink with words !!?)\nIn above , my aim is trying to calculate the dice to evaluate preformance of trained model , soi use the image in test folder to get the mask...\n(If my understanding is wrong , can you point out ? this is my first time to join kaggle...)\n\nThanks a lot!",
    "2201176": "I would like to make sure that the final predictions looks like the mask.png in test folder, rights? or mask.png in train folders or inklabels.png in train folders ? in 2D mask?",
    "2202019": "No, mask.png only shows which pixels contain data. The final predictions should look like inklabels.png.",
    "2202333": "i see ! Thanks for your replying! It is important to me！"
  },
  "source": "meta"
}