{
  "id": 69603,
  "title": "Question about CNN simple classifier",
  "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/69603",
  "author_name": "",
  "post_date": "2018-10-25T08:18:59.438315300Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Does anybody know if it is feasible to train a simple CNN to classify pneumonia or not? Using only the target values 0 and 1 and the images.</p>\n\n<p>I have tried a simple one but for now it assigns all labels to the class with more examples. (I guess it is a case of underfitting).</p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "410023",
      "postDate": "10/25/2018 08:18:59",
      "content": "<p>Does anybody know if it is feasible to train a simple CNN to classify pneumonia or not? Using only the target values 0 and 1 and the images.</p>\n\n<p>I have tried a simple one but for now it assigns all labels to the class with more examples. (I guess it is a case of underfitting).</p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "Does anybody know if it is feasible to train a simple CNN to classify pneumonia or not? Using only the target values 0 and 1 and the images.\n\nI have tried a simple one but for now it assigns all labels to the class with more examples. (I guess it is a case of underfitting).\n\nThanks!",
      "votes": null
    },
    {
      "id": "410024",
      "postDate": "10/25/2018 08:27:15",
      "content": "<p>Depends on what your definition of \"simple\" is. Chexnet is pretty decent and runs on u-net i think. </p>",
      "rawMarkdown": "Depends on what your definition of \"simple\" is. Chexnet is pretty decent and runs on u-net i think.",
      "votes": null
    },
    {
      "id": "410028",
      "postDate": "10/25/2018 08:37:06",
      "content": "<p>Yes, it works. \nI trained many classifiers(vgg/resnet/densenet....) without any problem.</p>",
      "rawMarkdown": "Yes, it works. \nI trained many classifiers(vgg/resnet/densenet....) without any problem.",
      "votes": null
    },
    {
      "id": "410248",
      "postDate": "10/25/2018 17:05:10",
      "content": "<p>Chexnet is based on a Densenet121 backbone (ie <a href=\"https://github.com/brucechou1983/CheXNet-Keras\">https://github.com/brucechou1983/CheXNet-Keras</a>)</p>",
      "rawMarkdown": "Chexnet is based on a Densenet121 backbone (ie https://github.com/brucechou1983/CheXNet-Keras)",
      "votes": null
    },
    {
      "id": "410315",
      "postDate": "10/25/2018 20:10:12",
      "content": "<p>Yes, you are right. Not really simple... I was on my phone and too lazy to check. Sorry! </p>",
      "rawMarkdown": "Yes, you are right. Not really simple... I was on my phone and too lazy to check. Sorry!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 410024,
      "author_name": "moshel",
      "author_url": "",
      "post_date": "10/25/2018 08:27:15",
      "content": "<p>Depends on what your definition of \"simple\" is. Chexnet is pretty decent and runs on u-net i think. </p>",
      "votes": null,
      "replies": [
        {
          "id": 410248,
          "author_name": "robrasmussen",
          "author_url": "",
          "post_date": "10/25/2018 17:05:10",
          "content": "<p>Chexnet is based on a Densenet121 backbone (ie <a href=\"https://github.com/brucechou1983/CheXNet-Keras\">https://github.com/brucechou1983/CheXNet-Keras</a>)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 410315,
          "author_name": "moshel",
          "author_url": "",
          "post_date": "10/25/2018 20:10:12",
          "content": "<p>Yes, you are right. Not really simple... I was on my phone and too lazy to check. Sorry! </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 410028,
      "author_name": "atom1231",
      "author_url": "",
      "post_date": "10/25/2018 08:37:06",
      "content": "<p>Yes, it works. \nI trained many classifiers(vgg/resnet/densenet....) without any problem.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "410023": "Does anybody know if it is feasible to train a simple CNN to classify pneumonia or not? Using only the target values 0 and 1 and the images.\n\nI have tried a simple one but for now it assigns all labels to the class with more examples. (I guess it is a case of underfitting).\n\nThanks!",
    "410024": "Depends on what your definition of \"simple\" is. Chexnet is pretty decent and runs on u-net i think.",
    "410028": "Yes, it works. \nI trained many classifiers(vgg/resnet/densenet....) without any problem.",
    "410248": "Chexnet is based on a Densenet121 backbone (ie https://github.com/brucechou1983/CheXNet-Keras)",
    "410315": "Yes, you are right. Not really simple... I was on my phone and too lazy to check. Sorry!"
  },
  "source": "meta"
}