{
  "id": 113167,
  "title": "How are you validating?",
  "url": "/competitions/3d-object-detection-for-autonomous-vehicles/discussion/113167",
  "author_name": "Pascal Pfeiffer",
  "post_date": "2019-10-17T12:48:52.340000",
  "votes": 0,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n\n<p>I finally managed to break the 0.001 barrier <em>sigh</em> (lots of wrong transformations and assumptions...) and I am wondering how you guys are validating locally. I am currently using 140 scenes for training and 40 scenes for validation. Especially in the low count groups such as animals I assume the difference from validation to the test set can be huge. </p>\n\n<p>Ilu</p>",
  "messages": [
    {
      "id": 651389,
      "postDate": "2019-10-17T12:48:52.340Z",
      "content": "<p>Hi everyone,</p>\n\n<p>I finally managed to break the 0.001 barrier <em>sigh</em> (lots of wrong transformations and assumptions...) and I am wondering how you guys are validating locally. I am currently using 140 scenes for training and 40 scenes for validation. Especially in the low count groups such as animals I assume the difference from validation to the test set can be huge. </p>\n\n<p>Ilu</p>",
      "rawMarkdown": "Hi everyone,\n\nI finally managed to break the 0.001 barrier *sigh* (lots of wrong transformations and assumptions...) and I am wondering how you guys are validating locally. I am currently using 140 scenes for training and 40 scenes for validation. Especially in the low count groups such as animals I assume the difference from validation to the test set can be huge. \n\nIlu"
    },
    {
      "id": 658690,
      "postDate": "2019-10-26T11:29:39.183Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 658690,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-10-26T11:29:39.183000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "651389": "Hi everyone,\n\nI finally managed to break the 0.001 barrier *sigh* (lots of wrong transformations and assumptions...) and I am wondering how you guys are validating locally. I am currently using 140 scenes for training and 40 scenes for validation. Especially in the low count groups such as animals I assume the difference from validation to the test set can be huge. \n\nIlu",
    "658690": ""
  }
}