{
  "id": 121090,
  "title": "Do we need regress the car brand ?",
  "url": "/competitions/pku-autonomous-driving/discussion/121090",
  "author_name": "",
  "post_date": "2019-12-11T01:57:50.537812300Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p><a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a> <br>\nThis base kernel show a good start result.  But it only regress a binary score for  there is existing a car .  Should we regress the car brands for 'brand classes'?  Is it neccessary to regress the car brands(like 'benz' or 'changcheng' ),then we can using the 3d_model size to get more accurate result?</p>\n\n<p>Is my thought direction correct? </p>",
  "messages": [
    {
      "id": "692184",
      "postDate": "12/11/2019 01:57:50",
      "content": "<p><a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a> <br>\nThis base kernel show a good start result.  But it only regress a binary score for  there is existing a car .  Should we regress the car brands for 'brand classes'?  Is it neccessary to regress the car brands(like 'benz' or 'changcheng' ),then we can using the 3d_model size to get more accurate result?</p>\n\n<p>Is my thought direction correct? </p>",
      "rawMarkdown": "[https://www.kaggle.com/hocop1/centernet-baseline](https://www.kaggle.com/hocop1/centernet-baseline)  \nThis base kernel show a good start result.  But it only regress a binary score for  there is existing a car .  Should we regress the car brands for 'brand classes'?  Is it neccessary to regress the car brands(like 'benz' or 'changcheng' ),then we can using the 3d_model size to get more accurate result?\n\nIs my thought direction correct?",
      "votes": null
    },
    {
      "id": "693159",
      "postDate": "12/12/2019 05:01:40",
      "content": "<p>I don't think so unless you have a pretty big network.</p>",
      "rawMarkdown": "I don't think so unless you have a pretty big network.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 693159,
      "author_name": "tonychenxyz",
      "author_url": "",
      "post_date": "12/12/2019 05:01:40",
      "content": "<p>I don't think so unless you have a pretty big network.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "692184": "[https://www.kaggle.com/hocop1/centernet-baseline](https://www.kaggle.com/hocop1/centernet-baseline)  \nThis base kernel show a good start result.  But it only regress a binary score for  there is existing a car .  Should we regress the car brands for 'brand classes'?  Is it neccessary to regress the car brands(like 'benz' or 'changcheng' ),then we can using the 3d_model size to get more accurate result?\n\nIs my thought direction correct?",
    "693159": "I don't think so unless you have a pretty big network."
  },
  "source": "meta"
}