{
  "id": 206534,
  "title": "which data augment is effective？",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/206534",
  "author_name": "",
  "post_date": "2020-12-25T06:07:34.077767100Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I try randomcrop、randomswap、colorjitt。。。。<br>\nThese are not effective。</p>\n<p>ensembles is not  effective</p>",
  "messages": [
    {
      "id": "1125853",
      "postDate": "12/25/2020 06:07:34",
      "content": "<p>I try randomcrop、randomswap、colorjitt。。。。<br>\nThese are not effective。</p>\n<p>ensembles is not  effective</p>",
      "rawMarkdown": "I try randomcrop、randomswap、colorjitt。。。。\nThese are not effective。\n\nensembles is not  effective",
      "votes": null
    },
    {
      "id": "1126415",
      "postDate": "12/25/2020 15:28:45",
      "content": "<blockquote>\n  <p>ensembles is not effective</p>\n</blockquote>\n<p>I also tried and i get worst results.</p>",
      "rawMarkdown": "> ensembles is not effective\n\nI also tried and i get worst results.",
      "votes": null
    },
    {
      "id": "1126489",
      "postDate": "12/25/2020 16:44:07",
      "content": "<p>Ensembles work only if each model \"say something new to the other\", otherwise it leads to redundancy and performance degradation.</p>",
      "rawMarkdown": "Ensembles work only if each model \"say something new to the other\", otherwise it leads to redundancy and performance degradation.",
      "votes": null
    },
    {
      "id": "1140433",
      "postDate": "01/06/2021 02:25:12",
      "content": "<p>I'm a beginner, but I would guess random rotations would be a useful augmentation.  I noticed a small number of x-rays are already rotated quite a bit while browsing the images just now, but maybe not enough in the training data to make predictions robust to rotated x-rays.</p>",
      "rawMarkdown": "I'm a beginner, but I would guess random rotations would be a useful augmentation.  I noticed a small number of x-rays are already rotated quite a bit while browsing the images just now, but maybe not enough in the training data to make predictions robust to rotated x-rays.",
      "votes": null
    },
    {
      "id": "1141922",
      "postDate": "01/07/2021 02:05:49",
      "content": "<p>Try to combine different models like ResNet+Densenet, or EfficientNet+ResNet in simble averaging</p>",
      "rawMarkdown": "Try to combine different models like ResNet+Densenet, or EfficientNet+ResNet in simble averaging",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1126415,
      "author_name": "hiramcho",
      "author_url": "",
      "post_date": "12/25/2020 15:28:45",
      "content": "<blockquote>\n  <p>ensembles is not effective</p>\n</blockquote>\n<p>I also tried and i get worst results.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1126489,
      "author_name": "",
      "author_url": "",
      "post_date": "12/25/2020 16:44:07",
      "content": "<p>Ensembles work only if each model \"say something new to the other\", otherwise it leads to redundancy and performance degradation.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1140433,
      "author_name": "swatson",
      "author_url": "",
      "post_date": "01/06/2021 02:25:12",
      "content": "<p>I'm a beginner, but I would guess random rotations would be a useful augmentation.  I noticed a small number of x-rays are already rotated quite a bit while browsing the images just now, but maybe not enough in the training data to make predictions robust to rotated x-rays.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1141922,
      "author_name": "",
      "author_url": "",
      "post_date": "01/07/2021 02:05:49",
      "content": "<p>Try to combine different models like ResNet+Densenet, or EfficientNet+ResNet in simble averaging</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1125853": "I try randomcrop、randomswap、colorjitt。。。。\nThese are not effective。\n\nensembles is not  effective",
    "1126415": "> ensembles is not effective\n\nI also tried and i get worst results.",
    "1126489": "Ensembles work only if each model \"say something new to the other\", otherwise it leads to redundancy and performance degradation.",
    "1140433": "I'm a beginner, but I would guess random rotations would be a useful augmentation.  I noticed a small number of x-rays are already rotated quite a bit while browsing the images just now, but maybe not enough in the training data to make predictions robust to rotated x-rays.",
    "1141922": "Try to combine different models like ResNet+Densenet, or EfficientNet+ResNet in simble averaging"
  },
  "source": "meta"
}