{
  "id": 65653,
  "title": "Which network can improve the accuracy of detection",
  "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/65653",
  "author_name": "",
  "post_date": "2018-09-13T09:00:51.809223100Z",
  "votes": 2,
  "comment_count": 8,
  "views": 0,
  "content": "<p>i try to use faster-rcnn-resnet50-coco ，i get low accuracy, Can someone give me advice？</p>",
  "messages": [
    {
      "id": "386631",
      "postDate": "09/13/2018 09:00:51",
      "content": "<p>i try to use faster-rcnn-resnet50-coco ，i get low accuracy, Can someone give me advice？</p>",
      "rawMarkdown": "i try to use faster-rcnn-resnet50-coco ，i get low accuracy, Can someone give me advice？",
      "votes": null
    },
    {
      "id": "386744",
      "postDate": "09/13/2018 13:42:33",
      "content": "<p>The coco data set is not related to the pneumonia data set. In this case, transfer learning from not provide performance improvements. Since you are using a pre-trained model, are you freezing the backbone? You can start training from a pre-trained model and it should be fine, but if your backbone is frozen, your network will not adapt to the chest x-rays.</p>",
      "rawMarkdown": "The coco data set is not related to the pneumonia data set. In this case, transfer learning from not provide performance improvements. Since you are using a pre-trained model, are you freezing the backbone? You can start training from a pre-trained model and it should be fine, but if your backbone is frozen, your network will not adapt to the chest x-rays.",
      "votes": null
    },
    {
      "id": "386797",
      "postDate": "09/13/2018 16:26:14",
      "content": "<p>Pretrained models are not much useful, they are useless in this competition. You can train a standard CNN model on ChexNet Xray 14 dataset, and then use that for the detection.</p>",
      "rawMarkdown": "Pretrained models are not much useful, they are useless in this competition. You can train a standard CNN model on ChexNet Xray 14 dataset, and then use that for the detection.",
      "votes": null
    },
    {
      "id": "386948",
      "postDate": "09/14/2018 02:38:36",
      "content": "<p>Thank you very much for your advice，i will try to do it。</p>",
      "rawMarkdown": "Thank you very much for your advice，i will try to do it。",
      "votes": null
    },
    {
      "id": "386949",
      "postDate": "09/14/2018 02:43:14",
      "content": "<p>Thank you very much.I use faster-rcnn-resnet-coco model.ckpt to fine-tune it,but get low accuracy.so i will try to use ChexNet,I hope I can get a good score。</p>",
      "rawMarkdown": "Thank you very much.I use faster-rcnn-resnet-coco model.ckpt to fine-tune it,but get low accuracy.so i will try to use ChexNet,I hope I can get a good score。",
      "votes": null
    },
    {
      "id": "386951",
      "postDate": "09/14/2018 02:51:31",
      "content": "<p>Hi Tomar,Can you tall me how can i get ChexNet Xray 14 dataset?</p>",
      "rawMarkdown": "Hi Tomar,Can you tall me how can i get ChexNet Xray 14 dataset?",
      "votes": null
    },
    {
      "id": "387003",
      "postDate": "09/14/2018 05:25:00",
      "content": "<p><a href=\"https://nihcc.app.box.com/v/ChestXray-NIHCC\">https://nihcc.app.box.com/v/ChestXray-NIHCC</a></p>",
      "rawMarkdown": "https://nihcc.app.box.com/v/ChestXray-NIHCC",
      "votes": null
    },
    {
      "id": "387133",
      "postDate": "09/14/2018 11:42:21",
      "content": "<p>Thank you very muck!</p>",
      "rawMarkdown": "Thank you very muck!",
      "votes": null
    },
    {
      "id": "391537",
      "postDate": "09/22/2018 01:42:25",
      "content": "<p>Looks like the NIH ChexNet XRay 14 dataset and models pretrained on it are not allowed as it overlaps with test data, your solution may be disqualified if based on it.</p>\n\n<p><a href=\"https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962\">https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962</a></p>",
      "rawMarkdown": "Looks like the NIH ChexNet XRay 14 dataset and models pretrained on it are not allowed as it overlaps with test data, your solution may be disqualified if based on it.\n\nhttps://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 386744,
      "author_name": "arpandhatt",
      "author_url": "",
      "post_date": "09/13/2018 13:42:33",
      "content": "<p>The coco data set is not related to the pneumonia data set. In this case, transfer learning from not provide performance improvements. Since you are using a pre-trained model, are you freezing the backbone? You can start training from a pre-trained model and it should be fine, but if your backbone is frozen, your network will not adapt to the chest x-rays.</p>",
      "votes": null,
      "replies": [
        {
          "id": 386949,
          "author_name": "jersonal",
          "author_url": "",
          "post_date": "09/14/2018 02:43:14",
          "content": "<p>Thank you very much.I use faster-rcnn-resnet-coco model.ckpt to fine-tune it,but get low accuracy.so i will try to use ChexNet,I hope I can get a good score。</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 386797,
      "author_name": "nikhilroxtomar",
      "author_url": "",
      "post_date": "09/13/2018 16:26:14",
      "content": "<p>Pretrained models are not much useful, they are useless in this competition. You can train a standard CNN model on ChexNet Xray 14 dataset, and then use that for the detection.</p>",
      "votes": null,
      "replies": [
        {
          "id": 386948,
          "author_name": "jersonal",
          "author_url": "",
          "post_date": "09/14/2018 02:38:36",
          "content": "<p>Thank you very much for your advice，i will try to do it。</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 386951,
          "author_name": "jersonal",
          "author_url": "",
          "post_date": "09/14/2018 02:51:31",
          "content": "<p>Hi Tomar,Can you tall me how can i get ChexNet Xray 14 dataset?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 387003,
          "author_name": "nikhilroxtomar",
          "author_url": "",
          "post_date": "09/14/2018 05:25:00",
          "content": "<p><a href=\"https://nihcc.app.box.com/v/ChestXray-NIHCC\">https://nihcc.app.box.com/v/ChestXray-NIHCC</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 387133,
          "author_name": "jersonal",
          "author_url": "",
          "post_date": "09/14/2018 11:42:21",
          "content": "<p>Thank you very muck!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 391537,
          "author_name": "dmytropoplavskiy",
          "author_url": "",
          "post_date": "09/22/2018 01:42:25",
          "content": "<p>Looks like the NIH ChexNet XRay 14 dataset and models pretrained on it are not allowed as it overlaps with test data, your solution may be disqualified if based on it.</p>\n\n<p><a href=\"https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962\">https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "386631": "i try to use faster-rcnn-resnet50-coco ，i get low accuracy, Can someone give me advice？",
    "386744": "The coco data set is not related to the pneumonia data set. In this case, transfer learning from not provide performance improvements. Since you are using a pre-trained model, are you freezing the backbone? You can start training from a pre-trained model and it should be fine, but if your backbone is frozen, your network will not adapt to the chest x-rays.",
    "386797": "Pretrained models are not much useful, they are useless in this competition. You can train a standard CNN model on ChexNet Xray 14 dataset, and then use that for the detection.",
    "386948": "Thank you very much for your advice，i will try to do it。",
    "386949": "Thank you very much.I use faster-rcnn-resnet-coco model.ckpt to fine-tune it,but get low accuracy.so i will try to use ChexNet,I hope I can get a good score。",
    "386951": "Hi Tomar,Can you tall me how can i get ChexNet Xray 14 dataset?",
    "387003": "https://nihcc.app.box.com/v/ChestXray-NIHCC",
    "387133": "Thank you very muck!",
    "391537": "Looks like the NIH ChexNet XRay 14 dataset and models pretrained on it are not allowed as it overlaps with test data, your solution may be disqualified if based on it.\n\nhttps://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/64345#388962"
  },
  "source": "meta"
}