{
  "id": 118361,
  "title": "Basic Model Creation - Reference Kernel ",
  "url": "/competitions/pku-autonomous-driving/discussion/118361",
  "author_name": "Nirjhar Roy",
  "post_date": "2019-11-21T02:04:01.859000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>After understanding few basic concepts as explained in my earlier topic . I have now tried understanding new model creation for this competition . One excellent base model and the only for now is <a href=\"/hocop1\">@hocop1</a>  kernel \n<a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a></p>\n\n<p>I have based my code on resnet encoder part as shown in the OFT code and kept the centernet portion and losses as per the public kernel .  Please refer this discussion for further info and code path \n<a href=\"https://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878\">https://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878</a></p>\n\n<p>For now I have used resnet18 encoder , you can modify for other encoder and experiment and let all of us know what is the result . </p>\n\n<p>As you can see the result are not yet very good (but not very bad also ) as compared to efficientnetb0 . And  it is also 2 times faster for same image and batchsize, so more epochs can be run .</p>\n\n<p>Here is the Kernel link \n<a href=\"https://www.kaggle.com/phoenix9032/center-resnet-trial\">https://www.kaggle.com/phoenix9032/center-resnet-trial</a></p>",
  "messages": [
    {
      "id": 678110,
      "postDate": "2019-11-21T02:04:01.860Z",
      "content": "<p>After understanding few basic concepts as explained in my earlier topic . I have now tried understanding new model creation for this competition . One excellent base model and the only for now is <a href=\"/hocop1\">@hocop1</a>  kernel \n<a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a></p>\n\n<p>I have based my code on resnet encoder part as shown in the OFT code and kept the centernet portion and losses as per the public kernel .  Please refer this discussion for further info and code path \n<a href=\"https://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878\">https://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878</a></p>\n\n<p>For now I have used resnet18 encoder , you can modify for other encoder and experiment and let all of us know what is the result . </p>\n\n<p>As you can see the result are not yet very good (but not very bad also ) as compared to efficientnetb0 . And  it is also 2 times faster for same image and batchsize, so more epochs can be run .</p>\n\n<p>Here is the Kernel link \n<a href=\"https://www.kaggle.com/phoenix9032/center-resnet-trial\">https://www.kaggle.com/phoenix9032/center-resnet-trial</a></p>",
      "rawMarkdown": "After understanding few basic concepts as explained in my earlier topic . I have now tried understanding new model creation for this competition . One excellent base model and the only for now is @hocop1  kernel \nhttps://www.kaggle.com/hocop1/centernet-baseline\n\nI have based my code on resnet encoder part as shown in the OFT code and kept the centernet portion and losses as per the public kernel .  Please refer this discussion for further info and code path \nhttps://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878\n\nFor now I have used resnet18 encoder , you can modify for other encoder and experiment and let all of us know what is the result . \n\nAs you can see the result are not yet very good (but not very bad also ) as compared to efficientnetb0 . And  it is also 2 times faster for same image and batchsize, so more epochs can be run .\n\nHere is the Kernel link \nhttps://www.kaggle.com/phoenix9032/center-resnet-trial",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "678110": "After understanding few basic concepts as explained in my earlier topic . I have now tried understanding new model creation for this competition . One excellent base model and the only for now is @hocop1  kernel \nhttps://www.kaggle.com/hocop1/centernet-baseline\n\nI have based my code on resnet encoder part as shown in the OFT code and kept the centernet portion and losses as per the public kernel .  Please refer this discussion for further info and code path \nhttps://www.kaggle.com/c/pku-autonomous-driving/discussion/118304#latest-677878\n\nFor now I have used resnet18 encoder , you can modify for other encoder and experiment and let all of us know what is the result . \n\nAs you can see the result are not yet very good (but not very bad also ) as compared to efficientnetb0 . And  it is also 2 times faster for same image and batchsize, so more epochs can be run .\n\nHere is the Kernel link \nhttps://www.kaggle.com/phoenix9032/center-resnet-trial"
  }
}