{
  "id": 77599,
  "title": "Landmark detection using fastai",
  "url": "/competitions/humpback-whale-identification/discussion/77599",
  "author_name": "",
  "post_date": "2019-01-14T17:56:49.853109100Z",
  "votes": 12,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have put together a notebook for <a href=\"https://github.com/radekosmulski/whale/blob/master/landmark_detection.ipynb\">landmark detection</a> (left and right tips and the notch).</p>\n\n<p>Only 300 train examples and a very simple model. Predictions are far from perfect but with this little data and that simple of a model that seems like an okay level of performance.</p>\n\n<p>Might be useful for horizontally aligning the fluke or for some other transformations, especially should the performance be improved via constructing a more elaborate model or adding data.</p>",
  "messages": [
    {
      "id": "455863",
      "postDate": "01/14/2019 17:56:49",
      "content": "<p>I have put together a notebook for <a href=\"https://github.com/radekosmulski/whale/blob/master/landmark_detection.ipynb\">landmark detection</a> (left and right tips and the notch).</p>\n\n<p>Only 300 train examples and a very simple model. Predictions are far from perfect but with this little data and that simple of a model that seems like an okay level of performance.</p>\n\n<p>Might be useful for horizontally aligning the fluke or for some other transformations, especially should the performance be improved via constructing a more elaborate model or adding data.</p>",
      "rawMarkdown": "I have put together a notebook for [landmark detection](https://github.com/radekosmulski/whale/blob/master/landmark_detection.ipynb) (left and right tips and the notch).\n\nOnly 300 train examples and a very simple model. Predictions are far from perfect but with this little data and that simple of a model that seems like an okay level of performance.\n\nMight be useful for horizontally aligning the fluke or for some other transformations, especially should the performance be improved via constructing a more elaborate model or adding data.",
      "votes": null
    },
    {
      "id": "459600",
      "postDate": "01/22/2019 04:02:48",
      "content": "<p>hi, redek. Thank you for your sharing. How did you label the landmark?</p>",
      "rawMarkdown": "hi, redek. Thank you for your sharing. How did you label the landmark?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 459600,
      "author_name": "roycezjq",
      "author_url": "",
      "post_date": "01/22/2019 04:02:48",
      "content": "<p>hi, redek. Thank you for your sharing. How did you label the landmark?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "455863": "I have put together a notebook for [landmark detection](https://github.com/radekosmulski/whale/blob/master/landmark_detection.ipynb) (left and right tips and the notch).\n\nOnly 300 train examples and a very simple model. Predictions are far from perfect but with this little data and that simple of a model that seems like an okay level of performance.\n\nMight be useful for horizontally aligning the fluke or for some other transformations, especially should the performance be improved via constructing a more elaborate model or adding data.",
    "459600": "hi, redek. Thank you for your sharing. How did you label the landmark?"
  },
  "source": "meta"
}