{
  "id": 240645,
  "title": "Are all public LB images available in test folder [ignoring the current problem]?",
  "url": "/competitions/siim-covid19-detection/discussion/240645",
  "author_name": "",
  "post_date": "2021-05-20T19:04:43.470452400Z",
  "votes": 5,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Ignoring the current issue with number of images in train and test folders and assuming it will be resolved soon, are we going to have all the images of public LB in test folder before submission? Or this folder is updated when we commit the notebook like most of other code competitions?</p>\n<p>Because it is strange for me that there are already about 1200 images in test so I think these are all LB test images (if they intended to show the structure of test folder, 5 or 10 images would suffice!) but on the other hand the internet must be off for submission which is usually when the test folder gets updated when submitting.</p>",
  "messages": [
    {
      "id": "1316668",
      "postDate": "05/20/2021 19:04:43",
      "content": "<p>Ignoring the current issue with number of images in train and test folders and assuming it will be resolved soon, are we going to have all the images of public LB in test folder before submission? Or this folder is updated when we commit the notebook like most of other code competitions?</p>\n<p>Because it is strange for me that there are already about 1200 images in test so I think these are all LB test images (if they intended to show the structure of test folder, 5 or 10 images would suffice!) but on the other hand the internet must be off for submission which is usually when the test folder gets updated when submitting.</p>",
      "rawMarkdown": "Ignoring the current issue with number of images in train and test folders and assuming it will be resolved soon, are we going to have all the images of public LB in test folder before submission? Or this folder is updated when we commit the notebook like most of other code competitions?\n\nBecause it is strange for me that there are already about 1200 images in test so I think these are all LB test images (if they intended to show the structure of test folder, 5 or 10 images would suffice!) but on the other hand the internet must be off for submission which is usually when the test folder gets updated when submitting.",
      "votes": null
    },
    {
      "id": "1316703",
      "postDate": "05/20/2021 19:48:27",
      "content": "<p>Yes, the test folder on the Data page contains all images scored on the public leaderboard. Because this is a code competition, at time of submission, your code is re-run synchronously on the private test set (swapping out the public test set for the private one) in the background. That score is withheld until the end of the competition. Note that your code is not considered submitted when you commit/save the notebook; you must submit that saved code/notebook to the competition. We keep the internet off on submission intentionally, as having it on would make it trivial to access the private test set.</p>",
      "rawMarkdown": "Yes, the test folder on the Data page contains all images scored on the public leaderboard. Because this is a code competition, at time of submission, your code is re-run synchronously on the private test set (swapping out the public test set for the private one) in the background. That score is withheld until the end of the competition. Note that your code is not considered submitted when you commit/save the notebook; you must submit that saved code/notebook to the competition. We keep the internet off on submission intentionally, as having it on would make it trivial to access the private test set.",
      "votes": null
    },
    {
      "id": "1316721",
      "postDate": "05/20/2021 20:04:12",
      "content": "<p>thanks <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> for the reply but it seems that I've complete all the procedures that you mentioned. But the score is very low and moreover, this is happening for other people too. So, there are two possibilites,</p>\n<ul>\n<li>There is something wrong with test labels, perhaps <strong>bbox format</strong> , in train data we have <strong>coco</strong> format but in submission we are asked to submit in <strong>voc_pascal</strong> format. Please check the test labels.</li>\n<li>The task is so difficult so difficult that a lot of people are struggling.</li>\n</ul>",
      "rawMarkdown": "thanks @juliaelliott for the reply but it seems that I've complete all the procedures that you mentioned. But the score is very low and moreover, this is happening for other people too. So, there are two possibilites,\n* There is something wrong with test labels, perhaps **bbox format** , in train data we have **coco** format but in submission we are asked to submit in **voc_pascal** format. Please check the test labels.\n* The task is so difficult so difficult that a lot of people are struggling.",
      "votes": null
    },
    {
      "id": "1316960",
      "postDate": "05/21/2021 04:18:08",
      "content": "<p>Thanks <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> for the clarification.</p>",
      "rawMarkdown": "Thanks @juliaelliott for the clarification.",
      "votes": null
    },
    {
      "id": "1317092",
      "postDate": "05/21/2021 06:45:43",
      "content": "<p>For image_label, BBox format \"xmin, ymin, xmax, ymax\" → \"xmin, ymin, w, h\" changed the LB score by only 0.001. <br>\nStrange…😂</p>",
      "rawMarkdown": "For image_label, BBox format \"xmin, ymin, xmax, ymax\" → \"xmin, ymin, w, h\" changed the LB score by only 0.001. \nStrange...😂",
      "votes": null
    },
    {
      "id": "1317682",
      "postDate": "05/21/2021 15:50:52",
      "content": "<p>I see you at the top of LB <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a> ! Did you solve the issue with submission? Image-Level Studies helped?</p>",
      "rawMarkdown": "I see you at the top of LB @awsaf49 ! Did you solve the issue with submission? Image-Level Studies helped?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1316703,
      "author_name": "juliaelliott",
      "author_url": "",
      "post_date": "05/20/2021 19:48:27",
      "content": "<p>Yes, the test folder on the Data page contains all images scored on the public leaderboard. Because this is a code competition, at time of submission, your code is re-run synchronously on the private test set (swapping out the public test set for the private one) in the background. That score is withheld until the end of the competition. Note that your code is not considered submitted when you commit/save the notebook; you must submit that saved code/notebook to the competition. We keep the internet off on submission intentionally, as having it on would make it trivial to access the private test set.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1316721,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "05/20/2021 20:04:12",
          "content": "<p>thanks <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> for the reply but it seems that I've complete all the procedures that you mentioned. But the score is very low and moreover, this is happening for other people too. So, there are two possibilites,</p>\n<ul>\n<li>There is something wrong with test labels, perhaps <strong>bbox format</strong> , in train data we have <strong>coco</strong> format but in submission we are asked to submit in <strong>voc_pascal</strong> format. Please check the test labels.</li>\n<li>The task is so difficult so difficult that a lot of people are struggling.</li>\n</ul>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1316960,
          "author_name": "moeinshariatnia",
          "author_url": "",
          "post_date": "05/21/2021 04:18:08",
          "content": "<p>Thanks <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> for the clarification.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1317092,
          "author_name": "drtausamaru",
          "author_url": "",
          "post_date": "05/21/2021 06:45:43",
          "content": "<p>For image_label, BBox format \"xmin, ymin, xmax, ymax\" → \"xmin, ymin, w, h\" changed the LB score by only 0.001. <br>\nStrange…😂</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1317682,
      "author_name": "moeinshariatnia",
      "author_url": "",
      "post_date": "05/21/2021 15:50:52",
      "content": "<p>I see you at the top of LB <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a> ! Did you solve the issue with submission? Image-Level Studies helped?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1316668": "Ignoring the current issue with number of images in train and test folders and assuming it will be resolved soon, are we going to have all the images of public LB in test folder before submission? Or this folder is updated when we commit the notebook like most of other code competitions?\n\nBecause it is strange for me that there are already about 1200 images in test so I think these are all LB test images (if they intended to show the structure of test folder, 5 or 10 images would suffice!) but on the other hand the internet must be off for submission which is usually when the test folder gets updated when submitting.",
    "1316703": "Yes, the test folder on the Data page contains all images scored on the public leaderboard. Because this is a code competition, at time of submission, your code is re-run synchronously on the private test set (swapping out the public test set for the private one) in the background. That score is withheld until the end of the competition. Note that your code is not considered submitted when you commit/save the notebook; you must submit that saved code/notebook to the competition. We keep the internet off on submission intentionally, as having it on would make it trivial to access the private test set.",
    "1316721": "thanks @juliaelliott for the reply but it seems that I've complete all the procedures that you mentioned. But the score is very low and moreover, this is happening for other people too. So, there are two possibilites,\n* There is something wrong with test labels, perhaps **bbox format** , in train data we have **coco** format but in submission we are asked to submit in **voc_pascal** format. Please check the test labels.\n* The task is so difficult so difficult that a lot of people are struggling.",
    "1316960": "Thanks @juliaelliott for the clarification.",
    "1317092": "For image_label, BBox format \"xmin, ymin, xmax, ymax\" → \"xmin, ymin, w, h\" changed the LB score by only 0.001. \nStrange...😂",
    "1317682": "I see you at the top of LB @awsaf49 ! Did you solve the issue with submission? Image-Level Studies helped?"
  },
  "source": "meta"
}