{
  "id": 234381,
  "title": "Problems with D4",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/234381",
  "author_name": "",
  "post_date": "2021-04-24T02:13:46.795784700Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I've been investigating my model's poor LB-CV gap. I have managed to boost it to 91% LB, and when investigating further, I found that the model performs well(93-94% LB) on all images except for d4. I know hand labelling exists, but for those who didn't hand label, how do we combat the issue of this one kidney?</p>",
  "messages": [
    {
      "id": "1282498",
      "postDate": "04/24/2021 02:13:46",
      "content": "<p>I've been investigating my model's poor LB-CV gap. I have managed to boost it to 91% LB, and when investigating further, I found that the model performs well(93-94% LB) on all images except for d4. I know hand labelling exists, but for those who didn't hand label, how do we combat the issue of this one kidney?</p>",
      "rawMarkdown": "I've been investigating my model's poor LB-CV gap. I have managed to boost it to 91% LB, and when investigating further, I found that the model performs well(93-94% LB) on all images except for d4. I know hand labelling exists, but for those who didn't hand label, how do we combat the issue of this one kidney?",
      "votes": null
    },
    {
      "id": "1283073",
      "postDate": "04/24/2021 14:55:11",
      "content": "<p>There is a topic with an attached dataset for hand labelled D48 labels <a href=\"https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/227616\" target=\"_blank\">here</a>.</p>\n<p>A lot of people, myself included, are using that. Nobody's quite sure if it's a good idea, or not, though…</p>",
      "rawMarkdown": "There is a topic with an attached dataset for hand labelled D48 labels [here](https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/227616).\n\nA lot of people, myself included, are using that. Nobody's quite sure if it's a good idea, or not, though...",
      "votes": null
    },
    {
      "id": "1283139",
      "postDate": "04/24/2021 15:59:03",
      "content": "<p>I'm just worried since my LB is much lower than other models without hand labels. I'm not sure how they improved that one kidney.</p>",
      "rawMarkdown": "I'm just worried since my LB is much lower than other models without hand labels. I'm not sure how they improved that one kidney.",
      "votes": null
    },
    {
      "id": "1283140",
      "postDate": "04/24/2021 16:02:57",
      "content": "<p>What was your LB without hand label, and if you don't mind sharing, how did you improve LB without switching to hand labels?</p>",
      "rawMarkdown": "What was your LB without hand label, and if you don't mind sharing, how did you improve LB without switching to hand labels?",
      "votes": null
    },
    {
      "id": "1283177",
      "postDate": "04/24/2021 16:45:38",
      "content": "<p>My single model went from 7.22 to 7.33 using the file linked above of the D48 hand label only.</p>",
      "rawMarkdown": "My single model went from 7.22 to 7.33 using the file linked above of the D48 hand label only.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1283073,
      "author_name": "jamesphoward",
      "author_url": "",
      "post_date": "04/24/2021 14:55:11",
      "content": "<p>There is a topic with an attached dataset for hand labelled D48 labels <a href=\"https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/227616\" target=\"_blank\">here</a>.</p>\n<p>A lot of people, myself included, are using that. Nobody's quite sure if it's a good idea, or not, though…</p>",
      "votes": null,
      "replies": [
        {
          "id": 1283139,
          "author_name": "andrewshao05",
          "author_url": "",
          "post_date": "04/24/2021 15:59:03",
          "content": "<p>I'm just worried since my LB is much lower than other models without hand labels. I'm not sure how they improved that one kidney.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1283140,
          "author_name": "andrewshao05",
          "author_url": "",
          "post_date": "04/24/2021 16:02:57",
          "content": "<p>What was your LB without hand label, and if you don't mind sharing, how did you improve LB without switching to hand labels?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1283177,
          "author_name": "jamesphoward",
          "author_url": "",
          "post_date": "04/24/2021 16:45:38",
          "content": "<p>My single model went from 7.22 to 7.33 using the file linked above of the D48 hand label only.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1282498": "I've been investigating my model's poor LB-CV gap. I have managed to boost it to 91% LB, and when investigating further, I found that the model performs well(93-94% LB) on all images except for d4. I know hand labelling exists, but for those who didn't hand label, how do we combat the issue of this one kidney?",
    "1283073": "There is a topic with an attached dataset for hand labelled D48 labels [here](https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/227616).\n\nA lot of people, myself included, are using that. Nobody's quite sure if it's a good idea, or not, though...",
    "1283139": "I'm just worried since my LB is much lower than other models without hand labels. I'm not sure how they improved that one kidney.",
    "1283140": "What was your LB without hand label, and if you don't mind sharing, how did you improve LB without switching to hand labels?",
    "1283177": "My single model went from 7.22 to 7.33 using the file linked above of the D48 hand label only."
  },
  "source": "meta"
}