{
  "id": 30290,
  "title": "imagenet pre-trained models",
  "url": "/competitions/intel-mobileodt-cervical-cancer-screening/discussion/30290",
  "author_name": "",
  "post_date": "2017-03-17T22:34:38.796746800Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Has anyone been able to fine tune a model with pre-trained imagenet weights on this data?</p>\n\n<p>I've tried a range of learning rates (0.0001  - 0.01) and also varied the number of trainable layers (i.e. just the top layer all the way through to training every layer). Xception and InceptionV3 in Keras (tf) haven't yielded any meaningful results.</p>\n\n<p>If you have any thoughts on hyper-parameters to try, I'm all ears.</p>",
  "messages": [
    {
      "id": "168726",
      "postDate": "03/17/2017 22:34:38",
      "content": "<p>Has anyone been able to fine tune a model with pre-trained imagenet weights on this data?</p>\n\n<p>I've tried a range of learning rates (0.0001  - 0.01) and also varied the number of trainable layers (i.e. just the top layer all the way through to training every layer). Xception and InceptionV3 in Keras (tf) haven't yielded any meaningful results.</p>\n\n<p>If you have any thoughts on hyper-parameters to try, I'm all ears.</p>",
      "rawMarkdown": "Has anyone been able to fine tune a model with pre-trained imagenet weights on this data?\n\nI've tried a range of learning rates (0.0001  - 0.01) and also varied the number of trainable layers (i.e. just the top layer all the way through to training every layer). Xception and InceptionV3 in Keras (tf) haven't yielded any meaningful results.\n\nIf you have any thoughts on hyper-parameters to try, I'm all ears.",
      "votes": null
    },
    {
      "id": "168746",
      "postDate": "03/17/2017 23:54:15",
      "content": "<p>i will try to use faster rcnn, but i need time to cut and label data. </p>",
      "rawMarkdown": "i will try to use faster rcnn, but i need time to cut and label data.",
      "votes": null
    },
    {
      "id": "182119",
      "postDate": "05/12/2017 13:12:26",
      "content": "<p>I 've tried it, and got no good result too. I think it might point that such as Xception, or Inception structure are trying to optimize the problem with high output-dimension rather than fine-grand classification. Just my thought ..</p>",
      "rawMarkdown": "I 've tried it, and got no good result too. I think it might point that such as Xception, or Inception structure are trying to optimize the problem with high output-dimension rather than fine-grand classification. Just my thought ..",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 168746,
      "author_name": "hagorms",
      "author_url": "",
      "post_date": "03/17/2017 23:54:15",
      "content": "<p>i will try to use faster rcnn, but i need time to cut and label data. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 182119,
      "author_name": "kentchun33333",
      "author_url": "",
      "post_date": "05/12/2017 13:12:26",
      "content": "<p>I 've tried it, and got no good result too. I think it might point that such as Xception, or Inception structure are trying to optimize the problem with high output-dimension rather than fine-grand classification. Just my thought ..</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "168726": "Has anyone been able to fine tune a model with pre-trained imagenet weights on this data?\n\nI've tried a range of learning rates (0.0001  - 0.01) and also varied the number of trainable layers (i.e. just the top layer all the way through to training every layer). Xception and InceptionV3 in Keras (tf) haven't yielded any meaningful results.\n\nIf you have any thoughts on hyper-parameters to try, I'm all ears.",
    "168746": "i will try to use faster rcnn, but i need time to cut and label data.",
    "182119": "I 've tried it, and got no good result too. I think it might point that such as Xception, or Inception structure are trying to optimize the problem with high output-dimension rather than fine-grand classification. Just my thought .."
  },
  "source": "meta"
}