{
  "id": 21573,
  "title": "Team up（Keras or Caffe or TensorFlow）",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/21573",
  "author_name": "",
  "post_date": "2016-06-10T13:02:24.293Z",
  "votes": null,
  "comment_count": 2,
  "views": 939,
  "content": "<p>Hi everyone&#65306;\nOur team use Keras VGG16 model witch hit the LB at 0.45, Caffe ResNet hit the LB at 0.56, and also Caffe gooGleNet at 0.51. All by just finetuning the model, No data augmentation or pre/post processing(except [X - meanRGB/meanBGR]) or ensamble. But we believe that we miss sth because We think finetuning can hit at least 0.25. So We want some people who hit the LB at 0.3 to join us. We'll apreciate it! Our team familiar with Keras TensorFlow Caffe. \nThx very much!</p>",
  "messages": [
    {
      "id": "123224",
      "postDate": "06/10/2016 13:02:24",
      "content": "<p>Hi everyone&#65306;\nOur team use Keras VGG16 model witch hit the LB at 0.45, Caffe ResNet hit the LB at 0.56, and also Caffe gooGleNet at 0.51. All by just finetuning the model, No data augmentation or pre/post processing(except [X - meanRGB/meanBGR]) or ensamble. But we believe that we miss sth because We think finetuning can hit at least 0.25. So We want some people who hit the LB at 0.3 to join us. We'll apreciate it! Our team familiar with Keras TensorFlow Caffe. \nThx very much!</p>",
      "rawMarkdown": "Hi everyone：\r\nOur team use Keras VGG16 model witch hit the LB at 0.45, Caffe ResNet hit the LB at 0.56, and also Caffe gooGleNet at 0.51. All by just finetuning the model, No data augmentation or pre/post processing(except [X - meanRGB/meanBGR]) or ensamble. But we believe that we miss sth because We think finetuning can hit at least 0.25. So We want some people who hit the LB at 0.3 to join us. We'll apreciate it! Our team familiar with Keras TensorFlow Caffe. \r\nThx very much!",
      "votes": null
    },
    {
      "id": "123241",
      "postDate": "06/10/2016 14:35:07",
      "content": "<p>why don't you try with the Keras code posted in the forum, it achieves ~0.25 LB.</p>",
      "rawMarkdown": "why don't you try with the Keras code posted in the forum, it achieves ~0.25 LB.",
      "votes": null
    },
    {
      "id": "123461",
      "postDate": "06/12/2016 02:27:31",
      "content": "<p>I got it! Thx @SecondPlan</p>",
      "rawMarkdown": "I got it! Thx @SecondPlan",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 123241,
      "author_name": "usixuz",
      "author_url": "",
      "post_date": "06/10/2016 14:35:07",
      "content": "<p>why don't you try with the Keras code posted in the forum, it achieves ~0.25 LB.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 123461,
      "author_name": "meanku",
      "author_url": "",
      "post_date": "06/12/2016 02:27:31",
      "content": "<p>I got it! Thx @SecondPlan</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "123224": "Hi everyone：\r\nOur team use Keras VGG16 model witch hit the LB at 0.45, Caffe ResNet hit the LB at 0.56, and also Caffe gooGleNet at 0.51. All by just finetuning the model, No data augmentation or pre/post processing(except [X - meanRGB/meanBGR]) or ensamble. But we believe that we miss sth because We think finetuning can hit at least 0.25. So We want some people who hit the LB at 0.3 to join us. We'll apreciate it! Our team familiar with Keras TensorFlow Caffe. \r\nThx very much!",
    "123241": "why don't you try with the Keras code posted in the forum, it achieves ~0.25 LB.",
    "123461": "I got it! Thx @SecondPlan"
  },
  "source": "meta"
}