{
  "id": 221159,
  "title": "Insight regarding chestxray14",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/221159",
  "author_name": "Virilo Tejedor Aguilera",
  "post_date": "2021-02-21T16:00:17.223000",
  "votes": 28,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I found this image in both the Chestxray14 and Ranzcr competition dataset.</p>\n<p><img src=\"https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg\" alt=\"\"><br>\nRanzcr competition: 1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg</p>\n<p><img src=\"https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/00013670_135.png\" alt=\"\"><br>\nChestxray14: 00013670_135.png</p>\n<p>Fortunately, this dataset has over 100,000 images, for competitors to manually tag the dataset in 23 days.</p>\n<p>And chestxray14 is not labeled for the catheter classification problem.</p>\n<p>Maybe you can take advantage of it training on this dataset with pseudo-labels.  More than training on any other external dataset.</p>\n<p>PS: <a href=\"https://www.kaggle.com/maggiemd\" target=\"_blank\">@maggiemd</a> <a href=\"https://www.kaggle.com/philculliton\" target=\"_blank\">@philculliton</a> I'm sorry I found it 😬 I tagged you to this post, just to keep you informed</p>",
  "messages": [
    {
      "id": 1212798,
      "postDate": "2021-02-21T16:00:17.223Z",
      "content": "<p>I found this image in both the Chestxray14 and Ranzcr competition dataset.</p>\n<p><img src=\"https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg\" alt=\"\"><br>\nRanzcr competition: 1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg</p>\n<p><img src=\"https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/00013670_135.png\" alt=\"\"><br>\nChestxray14: 00013670_135.png</p>\n<p>Fortunately, this dataset has over 100,000 images, for competitors to manually tag the dataset in 23 days.</p>\n<p>And chestxray14 is not labeled for the catheter classification problem.</p>\n<p>Maybe you can take advantage of it training on this dataset with pseudo-labels.  More than training on any other external dataset.</p>\n<p>PS: <a href=\"https://www.kaggle.com/maggiemd\" target=\"_blank\">@maggiemd</a> <a href=\"https://www.kaggle.com/philculliton\" target=\"_blank\">@philculliton</a> I'm sorry I found it 😬 I tagged you to this post, just to keep you informed</p>",
      "rawMarkdown": "I found this image in both the Chestxray14 and Ranzcr competition dataset.\n\n![](https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg)\nRanzcr competition: 1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg\n\n![](https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/00013670_135.png)\nChestxray14: 00013670_135.png\n\nFortunately, this dataset has over 100,000 images, for competitors to manually tag the dataset in 23 days.\n\nAnd chestxray14 is not labeled for the catheter classification problem.\n\nMaybe you can take advantage of it training on this dataset with pseudo-labels.  More than training on any other external dataset.\n\nPS: @maggiemd @philculliton I'm sorry I found it 😬 I tagged you to this post, just to keep you informed",
      "votes": 28
    },
    {
      "id": 1212872,
      "postDate": "2021-02-21T17:13:02.690Z",
      "content": "<p>Great find. Have you found other overlapping images between the two datasets? </p>",
      "rawMarkdown": "Great find. Have you found other overlapping images between the two datasets? ",
      "votes": 1,
      "replies": [
        {
          "id": 1212908,
          "postDate": "2021-02-21T18:04:26.810Z",
          "content": "<p>No, I was checking <a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/211515\" target=\"_blank\">this idea</a>. I was looking for segmentation masks that could be hurting the model.</p>\n<p>I found that the inference for these two ones had the almost same catheter mask, and came from different datasets.</p>\n<p>I think there must be a lot of repeated samples.</p>",
          "rawMarkdown": "No, I was checking [this idea](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/211515). I was looking for segmentation masks that could be hurting the model.\n\nI found that the inference for these two ones had the almost same catheter mask, and came from different datasets.\n\nI think there must be a lot of repeated samples.",
          "votes": 2
        },
        {
          "id": 1213140,
          "postDate": "2021-02-21T21:23:36.323Z",
          "content": "<p>Thanks for the details. 👌</p>",
          "rawMarkdown": "Thanks for the details. 👌",
          "votes": 1
        },
        {
          "id": 1213304,
          "postDate": "2021-02-22T03:26:10.147Z",
          "content": "<p>I agree, having external datasets would help the game + competitiveness. Have you done some analysis/ sterilization to prove that there are multiple repeated ones that can still maintain fairness in this competition? Maybe the host of this competition might be wanting to look for that.</p>",
          "rawMarkdown": "I agree, having external datasets would help the game + competitiveness. Have you done some analysis/ sterilization to prove that there are multiple repeated ones that can still maintain fairness in this competition? Maybe the host of this competition might be wanting to look for that."
        }
      ]
    },
    {
      "id": 1212808,
      "postDate": "2021-02-21T16:11:16.830Z",
      "content": "<p>I uploaded the images</p>",
      "rawMarkdown": "I uploaded the images\n\n"
    },
    {
      "id": 1217066,
      "postDate": "2021-02-24T18:23:58.303Z",
      "rawMarkdown": "",
      "votes": -4,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1212872,
      "author_name": "Yassine Alouini",
      "author_url": "",
      "post_date": "2021-02-21T17:13:02.690000",
      "content": "<p>Great find. Have you found other overlapping images between the two datasets? </p>",
      "votes": 1,
      "replies": [
        {
          "id": 1212908,
          "author_name": "Virilo Tejedor Aguilera",
          "author_url": "",
          "post_date": "2021-02-21T18:04:26.810000",
          "content": "<p>No, I was checking <a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/211515\" target=\"_blank\">this idea</a>. I was looking for segmentation masks that could be hurting the model.</p>\n<p>I found that the inference for these two ones had the almost same catheter mask, and came from different datasets.</p>\n<p>I think there must be a lot of repeated samples.</p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 1213140,
          "author_name": "Yassine Alouini",
          "author_url": "",
          "post_date": "2021-02-21T21:23:36.323000",
          "content": "<p>Thanks for the details. 👌</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1213304,
          "author_name": "Andy Jian Zhou",
          "author_url": "",
          "post_date": "2021-02-22T03:26:10.147000",
          "content": "<p>I agree, having external datasets would help the game + competitiveness. Have you done some analysis/ sterilization to prove that there are multiple repeated ones that can still maintain fairness in this competition? Maybe the host of this competition might be wanting to look for that.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1212808,
      "author_name": "Virilo Tejedor Aguilera",
      "author_url": "",
      "post_date": "2021-02-21T16:11:16.830000",
      "content": "<p>I uploaded the images</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1217066,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-02-24T18:23:58.303000",
      "content": "",
      "votes": -4,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1212798": "I found this image in both the Chestxray14 and Ranzcr competition dataset.\n\n![](https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg)\nRanzcr competition: 1.2.826.0.1.3680043.8.498.29881127584204492167872865603232889453.jpg\n\n![](https://raw.githubusercontent.com/virilo/kaggle-forums/master/images/chestxray14_ranzcr/00013670_135.png)\nChestxray14: 00013670_135.png\n\nFortunately, this dataset has over 100,000 images, for competitors to manually tag the dataset in 23 days.\n\nAnd chestxray14 is not labeled for the catheter classification problem.\n\nMaybe you can take advantage of it training on this dataset with pseudo-labels.  More than training on any other external dataset.\n\nPS: @maggiemd @philculliton I'm sorry I found it 😬 I tagged you to this post, just to keep you informed",
    "1212872": "Great find. Have you found other overlapping images between the two datasets? ",
    "1212808": "I uploaded the images\n\n",
    "1217066": ""
  }
}