{
  "id": 54918,
  "title": "Deformable Covolutions in the Autoencoder",
  "url": "/competitions/data-science-bowl-2018/discussion/54918",
  "author_name": "leo_gan",
  "post_date": "2018-04-19T17:07:13.661000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>The <strong>Deformable convolutions</strong> described in <a href=\"https://arxiv.org/pdf/1703.06211.pdf\">this paper</a>. I didn't find the examples of usage the deformable convolutions in the autoencoders. Hope, this example help with it. The standard convolutions use the box, unchangeable shape of the filters. </p>\n\n<p><a href=\"https://drive.google.com/open?id=1plBEXVNCcMYzvAerZZ9qHbVg2QV4Pi9x\">That's how the deformable convolution works.</a></p>\n\n<p>Here is the comparison of the deformable and the standard convolutions.</p>\n\n<p><a href=\"https://drive.google.com/open?id=1sJUFYiMyYISlGLeVefpWFdc6ar8Sv_NL\">Comparision of the deformable and the standard convolutions.</a></p>\n\n<p>The deformable convolutions learn the filter shapes and adjust shapes to the most frequent cases. I thought that would be helpful in this particular case when the nucleus have very similar shapes. I didn't run a whole bunch of the experiments to make sure the deformable convolutions work better in the most cases. In my tests, the deformable convolutions work at last on a par or a little bit better comparing with standard convolutions.</p>\n\n<p><a href=\"https://www.kaggle.com/leogan/deformable-covolutions-in-the-autoencoder\">The kernel is here.</a></p>",
  "messages": [
    {
      "id": 316693,
      "postDate": "2018-04-19T17:07:13.660Z",
      "content": "<p>The <strong>Deformable convolutions</strong> described in <a href=\"https://arxiv.org/pdf/1703.06211.pdf\">this paper</a>. I didn't find the examples of usage the deformable convolutions in the autoencoders. Hope, this example help with it. The standard convolutions use the box, unchangeable shape of the filters. </p>\n\n<p><a href=\"https://drive.google.com/open?id=1plBEXVNCcMYzvAerZZ9qHbVg2QV4Pi9x\">That's how the deformable convolution works.</a></p>\n\n<p>Here is the comparison of the deformable and the standard convolutions.</p>\n\n<p><a href=\"https://drive.google.com/open?id=1sJUFYiMyYISlGLeVefpWFdc6ar8Sv_NL\">Comparision of the deformable and the standard convolutions.</a></p>\n\n<p>The deformable convolutions learn the filter shapes and adjust shapes to the most frequent cases. I thought that would be helpful in this particular case when the nucleus have very similar shapes. I didn't run a whole bunch of the experiments to make sure the deformable convolutions work better in the most cases. In my tests, the deformable convolutions work at last on a par or a little bit better comparing with standard convolutions.</p>\n\n<p><a href=\"https://www.kaggle.com/leogan/deformable-covolutions-in-the-autoencoder\">The kernel is here.</a></p>",
      "rawMarkdown": "\nThe **Deformable convolutions** described in [this paper][1]. I didn't find the examples of usage the deformable convolutions in the autoencoders. Hope, this example help with it. The standard convolutions use the box, unchangeable shape of the filters. \n\n[That's how the deformable convolution works.][2]\n\nHere is the comparison of the deformable and the standard convolutions.\n\n[Comparision of the deformable and the standard convolutions.][3]\n\nThe deformable convolutions learn the filter shapes and adjust shapes to the most frequent cases. I thought that would be helpful in this particular case when the nucleus have very similar shapes. I didn't run a whole bunch of the experiments to make sure the deformable convolutions work better in the most cases. In my tests, the deformable convolutions work at last on a par or a little bit better comparing with standard convolutions.\n\n[The kernel is here.][4]\n\n\n  [1]: https://arxiv.org/pdf/1703.06211.pdf\n  [2]: https://drive.google.com/open?id=1plBEXVNCcMYzvAerZZ9qHbVg2QV4Pi9x\n  [3]: https://drive.google.com/open?id=1sJUFYiMyYISlGLeVefpWFdc6ar8Sv_NL\n  [4]: https://www.kaggle.com/leogan/deformable-covolutions-in-the-autoencoder",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "316693": "\nThe **Deformable convolutions** described in [this paper][1]. I didn't find the examples of usage the deformable convolutions in the autoencoders. Hope, this example help with it. The standard convolutions use the box, unchangeable shape of the filters. \n\n[That's how the deformable convolution works.][2]\n\nHere is the comparison of the deformable and the standard convolutions.\n\n[Comparision of the deformable and the standard convolutions.][3]\n\nThe deformable convolutions learn the filter shapes and adjust shapes to the most frequent cases. I thought that would be helpful in this particular case when the nucleus have very similar shapes. I didn't run a whole bunch of the experiments to make sure the deformable convolutions work better in the most cases. In my tests, the deformable convolutions work at last on a par or a little bit better comparing with standard convolutions.\n\n[The kernel is here.][4]\n\n\n  [1]: https://arxiv.org/pdf/1703.06211.pdf\n  [2]: https://drive.google.com/open?id=1plBEXVNCcMYzvAerZZ9qHbVg2QV4Pi9x\n  [3]: https://drive.google.com/open?id=1sJUFYiMyYISlGLeVefpWFdc6ar8Sv_NL\n  [4]: https://www.kaggle.com/leogan/deformable-covolutions-in-the-autoencoder"
  }
}