{
  "id": 240522,
  "title": "Do we need to predict a study level and a image level output for every ID?",
  "url": "/competitions/siim-covid19-detection/discussion/240522",
  "author_name": "",
  "post_date": "2021-05-20T07:05:12.101182500Z",
  "votes": 3,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi all ,<br>\nAfter going through the data , I can see that there are two types of labels available one is the classification level label i.e the study_level_label and other is the bbox labels i.e the image_level labels.<br>\nNow from my understanding the prediction of both are to be done separately because we have the ids suffixed by (study_level or image_level in the submission) , so my question whether we have to predict a study level output and a image_level output for every image?</p>",
  "messages": [
    {
      "id": "1315886",
      "postDate": "05/20/2021 07:05:12",
      "content": "<p>Hi all ,<br>\nAfter going through the data , I can see that there are two types of labels available one is the classification level label i.e the study_level_label and other is the bbox labels i.e the image_level labels.<br>\nNow from my understanding the prediction of both are to be done separately because we have the ids suffixed by (study_level or image_level in the submission) , so my question whether we have to predict a study level output and a image_level output for every image?</p>",
      "rawMarkdown": "Hi all ,\nAfter going through the data , I can see that there are two types of labels available one is the classification level label i.e the study_level_label and other is the bbox labels i.e the image_level labels.\nNow from my understanding the prediction of both are to be done separately because we have the ids suffixed by (study_level or image_level in the submission) , so my question whether we have to predict a study level output and a image_level output for every image?",
      "votes": null
    },
    {
      "id": "1315932",
      "postDate": "05/20/2021 07:44:25",
      "content": "<p>In the sample submission there are 1263 entries for image and 1214 for study so there is propably some overlap (multiple images per study) so you don't have to submit exactly two predictions per item I think.</p>",
      "rawMarkdown": "In the sample submission there are 1263 entries for image and 1214 for study so there is propably some overlap (multiple images per study) so you don't have to submit exactly two predictions per item I think.",
      "votes": null
    },
    {
      "id": "1316127",
      "postDate": "05/20/2021 10:10:01",
      "content": "<p>There are two different IDs in the submission file, ones with <strong>_study</strong> which are at study-level and ones with <strong>_image</strong> which are at image-level. So, we would have to do predictions at both study-level and image-level. You can find more details regarding the submission <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/240329\" target=\"_blank\">here</a>.</p>\n<p>To answer your question, yes, we would have to predict at study-level and at image-level for each image. But, a study can have multiple images so the prediction at study-level is combined prediction at study-level of all these images.</p>",
      "rawMarkdown": "There are two different IDs in the submission file, ones with **_study** which are at study-level and ones with **_image** which are at image-level. So, we would have to do predictions at both study-level and image-level. You can find more details regarding the submission [here](https://www.kaggle.com/c/siim-covid19-detection/discussion/240329).\n\nTo answer your question, yes, we would have to predict at study-level and at image-level for each image. But, a study can have multiple images so the prediction at study-level is combined prediction at study-level of all these images.",
      "votes": null
    },
    {
      "id": "1316132",
      "postDate": "05/20/2021 10:15:33",
      "content": "<p>Yes, you don't have to submit two predictions for every image but you would still have to predict at study-level and at image-level for every image.</p>",
      "rawMarkdown": "Yes, you don't have to submit two predictions for every image but you would still have to predict at study-level and at image-level for every image.",
      "votes": null
    },
    {
      "id": "1317086",
      "postDate": "05/21/2021 06:41:30",
      "content": "<p>Thanks for the reply </p>",
      "rawMarkdown": "Thanks for the reply",
      "votes": null
    },
    {
      "id": "1346075",
      "postDate": "06/12/2021 05:25:19",
      "content": "<p>I'm not sure if my interpretation is correct or not.</p>\n<p>As per my understanding, when we say that there can be more than one image per study. That means one study (for example a patient) has different kinds of images (for example different angles, different machine, different lighting condition). This means the disease detected will be the same for all the images, the opacity and bounding boxes could be different.</p>",
      "rawMarkdown": "I'm not sure if my interpretation is correct or not.\n\nAs per my understanding, when we say that there can be more than one image per study. That means one study (for example a patient) has different kinds of images (for example different angles, different machine, different lighting condition). This means the disease detected will be the same for all the images, the opacity and bounding boxes could be different.",
      "votes": null
    },
    {
      "id": "1346174",
      "postDate": "06/12/2021 07:04:54",
      "content": "<p>Yes you are right but there has been a bit of confusion about multiple images within the same study. There are many discussion threads and notebooks about this for example:<br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/siim-covid19-detection/discussion/240878</a><br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/dschettler8845/covid-detection-studies-with-multiple-images-viz</a></p>",
      "rawMarkdown": "Yes you are right but there has been a bit of confusion about multiple images within the same study. There are many discussion threads and notebooks about this for example:\n[https://www.kaggle.com/c/siim-covid19-detection/discussion/240878](url)\n[https://www.kaggle.com/dschettler8845/covid-detection-studies-with-multiple-images-viz](url)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1315932,
      "author_name": "simon111",
      "author_url": "",
      "post_date": "05/20/2021 07:44:25",
      "content": "<p>In the sample submission there are 1263 entries for image and 1214 for study so there is propably some overlap (multiple images per study) so you don't have to submit exactly two predictions per item I think.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1316132,
          "author_name": "hassiahk",
          "author_url": "",
          "post_date": "05/20/2021 10:15:33",
          "content": "<p>Yes, you don't have to submit two predictions for every image but you would still have to predict at study-level and at image-level for every image.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1316127,
      "author_name": "hassiahk",
      "author_url": "",
      "post_date": "05/20/2021 10:10:01",
      "content": "<p>There are two different IDs in the submission file, ones with <strong>_study</strong> which are at study-level and ones with <strong>_image</strong> which are at image-level. So, we would have to do predictions at both study-level and image-level. You can find more details regarding the submission <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/240329\" target=\"_blank\">here</a>.</p>\n<p>To answer your question, yes, we would have to predict at study-level and at image-level for each image. But, a study can have multiple images so the prediction at study-level is combined prediction at study-level of all these images.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1317086,
          "author_name": "tanulsingh077",
          "author_url": "",
          "post_date": "05/21/2021 06:41:30",
          "content": "<p>Thanks for the reply </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1346075,
      "author_name": "harshalgarg",
      "author_url": "",
      "post_date": "06/12/2021 05:25:19",
      "content": "<p>I'm not sure if my interpretation is correct or not.</p>\n<p>As per my understanding, when we say that there can be more than one image per study. That means one study (for example a patient) has different kinds of images (for example different angles, different machine, different lighting condition). This means the disease detected will be the same for all the images, the opacity and bounding boxes could be different.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1346174,
          "author_name": "simon111",
          "author_url": "",
          "post_date": "06/12/2021 07:04:54",
          "content": "<p>Yes you are right but there has been a bit of confusion about multiple images within the same study. There are many discussion threads and notebooks about this for example:<br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/siim-covid19-detection/discussion/240878</a><br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/dschettler8845/covid-detection-studies-with-multiple-images-viz</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1315886": "Hi all ,\nAfter going through the data , I can see that there are two types of labels available one is the classification level label i.e the study_level_label and other is the bbox labels i.e the image_level labels.\nNow from my understanding the prediction of both are to be done separately because we have the ids suffixed by (study_level or image_level in the submission) , so my question whether we have to predict a study level output and a image_level output for every image?",
    "1315932": "In the sample submission there are 1263 entries for image and 1214 for study so there is propably some overlap (multiple images per study) so you don't have to submit exactly two predictions per item I think.",
    "1316127": "There are two different IDs in the submission file, ones with **_study** which are at study-level and ones with **_image** which are at image-level. So, we would have to do predictions at both study-level and image-level. You can find more details regarding the submission [here](https://www.kaggle.com/c/siim-covid19-detection/discussion/240329).\n\nTo answer your question, yes, we would have to predict at study-level and at image-level for each image. But, a study can have multiple images so the prediction at study-level is combined prediction at study-level of all these images.",
    "1316132": "Yes, you don't have to submit two predictions for every image but you would still have to predict at study-level and at image-level for every image.",
    "1317086": "Thanks for the reply",
    "1346075": "I'm not sure if my interpretation is correct or not.\n\nAs per my understanding, when we say that there can be more than one image per study. That means one study (for example a patient) has different kinds of images (for example different angles, different machine, different lighting condition). This means the disease detected will be the same for all the images, the opacity and bounding boxes could be different.",
    "1346174": "Yes you are right but there has been a bit of confusion about multiple images within the same study. There are many discussion threads and notebooks about this for example:\n[https://www.kaggle.com/c/siim-covid19-detection/discussion/240878](url)\n[https://www.kaggle.com/dschettler8845/covid-detection-studies-with-multiple-images-viz](url)"
  },
  "source": "meta"
}