{
  "id": 241408,
  "title": "Why object detection?",
  "url": "/competitions/siim-covid19-detection/discussion/241408",
  "author_name": "",
  "post_date": "2021-05-24T12:59:04.778608600Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>In the interests of stimulating discussion, I wanted to ask this question to the Kaggle community.</p>\n<p>What additional benefit is provided by object detection here?</p>\n<ul>\n<li>The end user is presumably a clinician who will use the classification output to determine scale of treatment.</li>\n<li>The opacities for a sufficiently severe case should be apparent to the same clinician. </li>\n<li>Is there some reasoning from the data science point of view, why this object detection information would be useful?</li>\n</ul>",
  "messages": [
    {
      "id": "1320964",
      "postDate": "05/24/2021 12:59:04",
      "content": "<p>In the interests of stimulating discussion, I wanted to ask this question to the Kaggle community.</p>\n<p>What additional benefit is provided by object detection here?</p>\n<ul>\n<li>The end user is presumably a clinician who will use the classification output to determine scale of treatment.</li>\n<li>The opacities for a sufficiently severe case should be apparent to the same clinician. </li>\n<li>Is there some reasoning from the data science point of view, why this object detection information would be useful?</li>\n</ul>",
      "rawMarkdown": "In the interests of stimulating discussion, I wanted to ask this question to the Kaggle community.\n\nWhat additional benefit is provided by object detection here?\n- The end user is presumably a clinician who will use the classification output to determine scale of treatment.\n- The opacities for a sufficiently severe case should be apparent to the same clinician. \n- Is there some reasoning from the data science point of view, why this object detection information would be useful?",
      "votes": null
    },
    {
      "id": "1321680",
      "postDate": "05/24/2021 21:11:55",
      "content": "<p>My personal take in favor of object detection over image classification:</p>\n<ul>\n<li>quantification of pathology (severity, follow up)</li>\n<li>objective and measurable spatial and anatomical explainability of positive findings</li>\n<li>easier to detect performance shift in real world context</li>\n</ul>\n<p>And against semantic or instance segmentation:</p>\n<ul>\n<li>Faster, easier and arguably more reproducible annotations by radiologists</li>\n</ul>",
      "rawMarkdown": "My personal take in favor of object detection over image classification:\n- quantification of pathology (severity, follow up)\n- objective and measurable spatial and anatomical explainability of positive findings\n- easier to detect performance shift in real world context\n\nAnd against semantic or instance segmentation:\n- Faster, easier and arguably more reproducible annotations by radiologists",
      "votes": null
    },
    {
      "id": "1322918",
      "postDate": "05/25/2021 19:35:22",
      "content": "<p>Thanks Alexandre for that great answer. Makes perfect sense.</p>",
      "rawMarkdown": "Thanks Alexandre for that great answer. Makes perfect sense.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1321680,
      "author_name": "alexandrecc",
      "author_url": "",
      "post_date": "05/24/2021 21:11:55",
      "content": "<p>My personal take in favor of object detection over image classification:</p>\n<ul>\n<li>quantification of pathology (severity, follow up)</li>\n<li>objective and measurable spatial and anatomical explainability of positive findings</li>\n<li>easier to detect performance shift in real world context</li>\n</ul>\n<p>And against semantic or instance segmentation:</p>\n<ul>\n<li>Faster, easier and arguably more reproducible annotations by radiologists</li>\n</ul>",
      "votes": null,
      "replies": [
        {
          "id": 1322918,
          "author_name": "reubenschmidt",
          "author_url": "",
          "post_date": "05/25/2021 19:35:22",
          "content": "<p>Thanks Alexandre for that great answer. Makes perfect sense.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1320964": "In the interests of stimulating discussion, I wanted to ask this question to the Kaggle community.\n\nWhat additional benefit is provided by object detection here?\n- The end user is presumably a clinician who will use the classification output to determine scale of treatment.\n- The opacities for a sufficiently severe case should be apparent to the same clinician. \n- Is there some reasoning from the data science point of view, why this object detection information would be useful?",
    "1321680": "My personal take in favor of object detection over image classification:\n- quantification of pathology (severity, follow up)\n- objective and measurable spatial and anatomical explainability of positive findings\n- easier to detect performance shift in real world context\n\nAnd against semantic or instance segmentation:\n- Faster, easier and arguably more reproducible annotations by radiologists",
    "1322918": "Thanks Alexandre for that great answer. Makes perfect sense."
  },
  "source": "meta"
}