{
  "id": 218749,
  "title": "Which image size is the best?",
  "url": "/competitions/vinbigdata-chest-xray-abnormalities-detection/discussion/218749",
  "author_name": "Shichang Liu",
  "post_date": "2021-02-12T01:27:34.613000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I notice some people use 256, 512 or 1024 version of this dataset, do they work better than the original dataset? I can train my model at original size, but don't know is other dataset worth to try?</p>",
  "messages": [
    {
      "id": 1197548,
      "postDate": "2021-02-12T08:34:02.803Z",
      "content": "<p>In my opinion, bigger size is better for performance. <br>\nAnd If you use small size image, you need to handle <code>small object issue</code> in some case.<br>\nBut of course there is trade-off if you use bigger size image - resource and time<br>\ntime means <code>Doing some experimental things</code> for me :) </p>",
      "rawMarkdown": "In my opinion, bigger size is better for performance. \nAnd If you use small size image, you need to handle `small object issue` in some case.\nBut of course there is trade-off if you use bigger size image - resource and time\ntime means `Doing some experimental things ` for me :) ",
      "votes": 1
    },
    {
      "id": 1197164,
      "postDate": "2021-02-12T01:27:34.613Z",
      "content": "<p>I notice some people use 256, 512 or 1024 version of this dataset, do they work better than the original dataset? I can train my model at original size, but don't know is other dataset worth to try?</p>",
      "rawMarkdown": "I notice some people use 256, 512 or 1024 version of this dataset, do they work better than the original dataset? I can train my model at original size, but don't know is other dataset worth to try?",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 1197548,
      "author_name": "JIN",
      "author_url": "",
      "post_date": "2021-02-12T08:34:02.803000",
      "content": "<p>In my opinion, bigger size is better for performance. <br>\nAnd If you use small size image, you need to handle <code>small object issue</code> in some case.<br>\nBut of course there is trade-off if you use bigger size image - resource and time<br>\ntime means <code>Doing some experimental things</code> for me :) </p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1197548": "In my opinion, bigger size is better for performance. \nAnd If you use small size image, you need to handle `small object issue` in some case.\nBut of course there is trade-off if you use bigger size image - resource and time\ntime means `Doing some experimental things ` for me :) ",
    "1197164": "I notice some people use 256, 512 or 1024 version of this dataset, do they work better than the original dataset? I can train my model at original size, but don't know is other dataset worth to try?"
  }
}