{
  "id": 218646,
  "title": "Roots Images",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/218646",
  "author_name": "",
  "post_date": "2021-02-11T11:49:28.593528900Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p><img src=\"https://image.prntscr.com/image/nSAXZgvbQjOcxmXmkWwmKw.png\" alt=\"https://prnt.sc/z35v4r\"></p>\n<p>I selected manually images containing roots from the dataset. It seems like almost  all of them are part of CBSD ( label 1).<br>\nIsn't there a problem? ( not having representative images for healthy roots).</p>",
  "messages": [
    {
      "id": "1196400",
      "postDate": "02/11/2021 11:49:28",
      "content": "<p><img src=\"https://image.prntscr.com/image/nSAXZgvbQjOcxmXmkWwmKw.png\" alt=\"https://prnt.sc/z35v4r\"></p>\n<p>I selected manually images containing roots from the dataset. It seems like almost  all of them are part of CBSD ( label 1).<br>\nIsn't there a problem? ( not having representative images for healthy roots).</p>",
      "rawMarkdown": "![https://prnt.sc/z35v4r](https://image.prntscr.com/image/nSAXZgvbQjOcxmXmkWwmKw.png)\n\nI selected manually images containing roots from the dataset. It seems like almost  all of them are part of CBSD ( label 1).\nIsn't there a problem? ( not having representative images for healthy roots).",
      "votes": null
    },
    {
      "id": "1196656",
      "postDate": "02/11/2021 14:31:09",
      "content": "<p>From this <a href=\"https://www.kaggle.com/bjoernholzhauer/cassava-leaf-disease-classif-eda-cv-strategy#Plotting-example-images-for-each-cluster\" target=\"_blank\">EDA notebook</a>, I thought the breakdown is something like:</p>\n<ul>\n<li>Cassava Mosaic Disease (CMD)           40</li>\n<li>Cassava Green Mottle (CGM)              9</li>\n<li>Cassava Brown Streak Disease (CBSD)     8</li>\n<li>Healthy                                 5</li>\n<li>Cassava Bacterial Blight (CBB)          4</li>\n</ul>\n<p>And, yes, very few examples of some of the classes is a bit of a problem. Data augmentation can of course help to an extent, but that's still tricky. Of course, it could be that we mostly have root images for one particular disease, because people mostly take images of roots when that disease is present. That might work on the test set, too? But hard to guess, the best solution would of course be to find more data…</p>",
      "rawMarkdown": "From this [EDA notebook](https://www.kaggle.com/bjoernholzhauer/cassava-leaf-disease-classif-eda-cv-strategy#Plotting-example-images-for-each-cluster), I thought the breakdown is something like:\n* Cassava Mosaic Disease (CMD)           40\n* Cassava Green Mottle (CGM)              9\n* Cassava Brown Streak Disease (CBSD)     8\n* Healthy                                 5\n* Cassava Bacterial Blight (CBB)          4\n\nAnd, yes, very few examples of some of the classes is a bit of a problem. Data augmentation can of course help to an extent, but that's still tricky. Of course, it could be that we mostly have root images for one particular disease, because people mostly take images of roots when that disease is present. That might work on the test set, too? But hard to guess, the best solution would of course be to find more data...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1196656,
      "author_name": "bjoernholzhauer",
      "author_url": "",
      "post_date": "02/11/2021 14:31:09",
      "content": "<p>From this <a href=\"https://www.kaggle.com/bjoernholzhauer/cassava-leaf-disease-classif-eda-cv-strategy#Plotting-example-images-for-each-cluster\" target=\"_blank\">EDA notebook</a>, I thought the breakdown is something like:</p>\n<ul>\n<li>Cassava Mosaic Disease (CMD)           40</li>\n<li>Cassava Green Mottle (CGM)              9</li>\n<li>Cassava Brown Streak Disease (CBSD)     8</li>\n<li>Healthy                                 5</li>\n<li>Cassava Bacterial Blight (CBB)          4</li>\n</ul>\n<p>And, yes, very few examples of some of the classes is a bit of a problem. Data augmentation can of course help to an extent, but that's still tricky. Of course, it could be that we mostly have root images for one particular disease, because people mostly take images of roots when that disease is present. That might work on the test set, too? But hard to guess, the best solution would of course be to find more data…</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1196400": "![https://prnt.sc/z35v4r](https://image.prntscr.com/image/nSAXZgvbQjOcxmXmkWwmKw.png)\n\nI selected manually images containing roots from the dataset. It seems like almost  all of them are part of CBSD ( label 1).\nIsn't there a problem? ( not having representative images for healthy roots).",
    "1196656": "From this [EDA notebook](https://www.kaggle.com/bjoernholzhauer/cassava-leaf-disease-classif-eda-cv-strategy#Plotting-example-images-for-each-cluster), I thought the breakdown is something like:\n* Cassava Mosaic Disease (CMD)           40\n* Cassava Green Mottle (CGM)              9\n* Cassava Brown Streak Disease (CBSD)     8\n* Healthy                                 5\n* Cassava Bacterial Blight (CBB)          4\n\nAnd, yes, very few examples of some of the classes is a bit of a problem. Data augmentation can of course help to an extent, but that's still tricky. Of course, it could be that we mostly have root images for one particular disease, because people mostly take images of roots when that disease is present. That might work on the test set, too? But hard to guess, the best solution would of course be to find more data..."
  },
  "source": "meta"
}