{
  "id": 199102,
  "title": "Some images are misclassified as Healthy",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/199102",
  "author_name": "",
  "post_date": "2020-11-24T11:57:55.937168900Z",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I can see Many images are mislabeled in different categories. for eg.., Some unhealthy images are marked healthy and some healthy ones are marked in other categories. see attachment</p>\n<p>How did you overcome this?</p>",
  "messages": [
    {
      "id": "1089327",
      "postDate": "11/24/2020 11:57:55",
      "content": "<p>I can see Many images are mislabeled in different categories. for eg.., Some unhealthy images are marked healthy and some healthy ones are marked in other categories. see attachment</p>\n<p>How did you overcome this?</p>",
      "rawMarkdown": "I can see Many images are mislabeled in different categories. for eg.., Some unhealthy images are marked healthy and some healthy ones are marked in other categories. see attachment\n\nHow did you overcome this?",
      "votes": null
    },
    {
      "id": "1096644",
      "postDate": "11/30/2020 16:56:51",
      "content": "<p>Not sure why no one answered this question. From what i see in the image you shared, it did look like it's misclassified. Hope someone responds!</p>",
      "rawMarkdown": "Not sure why no one answered this question. From what i see in the image you shared, it did look like it's misclassified. Hope someone responds!",
      "votes": null
    },
    {
      "id": "1097626",
      "postDate": "12/01/2020 06:21:09",
      "content": "<p>Probably a labeling error, one way of dealing with this is to train a model to get predictions on full training data. Later based on predictions you can calculate loss and plot images with highest loss. Then, you can manually exclude, skip or correct the labels for these images. I don't know if it would worth the effort though, can't know without trying.</p>",
      "rawMarkdown": "Probably a labeling error, one way of dealing with this is to train a model to get predictions on full training data. Later based on predictions you can calculate loss and plot images with highest loss. Then, you can manually exclude, skip or correct the labels for these images. I don't know if it would worth the effort though, can't know without trying.",
      "votes": null
    },
    {
      "id": "1100956",
      "postDate": "12/03/2020 14:02:15",
      "content": "<p>I think it worth a try. Because those indicates the outliers. maybe there are more. just a thought!</p>",
      "rawMarkdown": "I think it worth a try. Because those indicates the outliers. maybe there are more. just a thought!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1096644,
      "author_name": "rawwar",
      "author_url": "",
      "post_date": "11/30/2020 16:56:51",
      "content": "<p>Not sure why no one answered this question. From what i see in the image you shared, it did look like it's misclassified. Hope someone responds!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1097626,
      "author_name": "keremt",
      "author_url": "",
      "post_date": "12/01/2020 06:21:09",
      "content": "<p>Probably a labeling error, one way of dealing with this is to train a model to get predictions on full training data. Later based on predictions you can calculate loss and plot images with highest loss. Then, you can manually exclude, skip or correct the labels for these images. I don't know if it would worth the effort though, can't know without trying.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1100956,
          "author_name": "bmanikan",
          "author_url": "",
          "post_date": "12/03/2020 14:02:15",
          "content": "<p>I think it worth a try. Because those indicates the outliers. maybe there are more. just a thought!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1089327": "I can see Many images are mislabeled in different categories. for eg.., Some unhealthy images are marked healthy and some healthy ones are marked in other categories. see attachment\n\nHow did you overcome this?",
    "1096644": "Not sure why no one answered this question. From what i see in the image you shared, it did look like it's misclassified. Hope someone responds!",
    "1097626": "Probably a labeling error, one way of dealing with this is to train a model to get predictions on full training data. Later based on predictions you can calculate loss and plot images with highest loss. Then, you can manually exclude, skip or correct the labels for these images. I don't know if it would worth the effort though, can't know without trying.",
    "1100956": "I think it worth a try. Because those indicates the outliers. maybe there are more. just a thought!"
  },
  "source": "meta"
}