{
  "id": 154683,
  "title": "1st place solution in ISIC 2019 challenge (w/code)",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/154683",
  "author_name": "alxndrkalinin",
  "post_date": "2020-05-29T11:21:08.052000",
  "votes": 174,
  "comment_count": 23,
  "views": 0,
  "content": "<p>ISIC 2019 challenge 1st place solution <a href=\"https://doi.org/10.1016/j.mex.2020.100864\">paper</a> and <a href=\"https://github.com/ngessert/isic2019\">PyTorch code</a></p>\n\n<p>Key insights:\n- preprocessing: cropping + Shades of Gray color constancy\n-  EfficientNets (<a href=\"https://github.com/lukemelas/EfficientNet-PyTorch\">EfficientNet-PyTorch</a>) + SENet154\n- extensive augs + CutOut\n- ensembling that maximizes the mean sensitivity\n- MLP for meta-data</p>\n\n<p><img src=\"https://ars.els-cdn.com/content/image/1-s2.0-S2215016120300832-gr2.jpg\" alt=\"network architecture\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F66669%2Fb08311fcc713bc4796fe24ab28872b93%2FScreen%20Shot%202020-05-29%20at%201.25.47%20PM.png?generation=1590750836068470&amp;alt=media\" alt=\"\"></p>\n\n<p><a href=\"https://challenge2019.isic-archive.com/leaderboard.html\">ISIC 2019 leaderboard</a> contains links to papers describing various other approaches. E.g. 3rd place built  an ensemble of leave out classifiers.</p>",
  "messages": [
    {
      "id": 866410,
      "postDate": "2020-05-29T11:21:08.053Z",
      "content": "<p>ISIC 2019 challenge 1st place solution <a href=\"https://doi.org/10.1016/j.mex.2020.100864\">paper</a> and <a href=\"https://github.com/ngessert/isic2019\">PyTorch code</a></p>\n\n<p>Key insights:\n- preprocessing: cropping + Shades of Gray color constancy\n-  EfficientNets (<a href=\"https://github.com/lukemelas/EfficientNet-PyTorch\">EfficientNet-PyTorch</a>) + SENet154\n- extensive augs + CutOut\n- ensembling that maximizes the mean sensitivity\n- MLP for meta-data</p>\n\n<p><img src=\"https://ars.els-cdn.com/content/image/1-s2.0-S2215016120300832-gr2.jpg\" alt=\"network architecture\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F66669%2Fb08311fcc713bc4796fe24ab28872b93%2FScreen%20Shot%202020-05-29%20at%201.25.47%20PM.png?generation=1590750836068470&amp;alt=media\" alt=\"\"></p>\n\n<p><a href=\"https://challenge2019.isic-archive.com/leaderboard.html\">ISIC 2019 leaderboard</a> contains links to papers describing various other approaches. E.g. 3rd place built  an ensemble of leave out classifiers.</p>",
      "rawMarkdown": "ISIC 2019 challenge 1st place solution [paper](https://doi.org/10.1016/j.mex.2020.100864) and [PyTorch code](https://github.com/ngessert/isic2019)\n\nKey insights:\n- preprocessing: cropping + Shades of Gray color constancy\n-  EfficientNets ([EfficientNet-PyTorch](https://github.com/lukemelas/EfficientNet-PyTorch)) + SENet154\n- extensive augs + CutOut\n- ensembling that maximizes the mean sensitivity\n- MLP for meta-data\n\n![network architecture](https://ars.els-cdn.com/content/image/1-s2.0-S2215016120300832-gr2.jpg)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F66669%2Fb08311fcc713bc4796fe24ab28872b93%2FScreen%20Shot%202020-05-29%20at%201.25.47%20PM.png?generation=1590750836068470&amp;alt=media)\n\n\n[ISIC 2019 leaderboard](https://challenge2019.isic-archive.com/leaderboard.html) contains links to papers describing various other approaches. E.g. 3rd place built  an ensemble of leave out classifiers.\n",
      "votes": 173
    },
    {
      "id": 969631,
      "postDate": "2020-08-13T20:10:58.090Z",
      "content": "<p>Nice work!</p>",
      "rawMarkdown": "Nice work!",
      "votes": 1
    },
    {
      "id": 963672,
      "postDate": "2020-08-09T07:46:11.190Z",
      "content": "<p>Thanks for sharing! helped me learn a lot form the ay you picked up insights from the data.</p>",
      "rawMarkdown": "Thanks for sharing! helped me learn a lot form the ay you picked up insights from the data.\n",
      "votes": 1
    },
    {
      "id": 893542,
      "postDate": "2020-06-19T17:51:34.467Z",
      "content": "<p>Thank you for sharing. I am gonna try to implement it with DenseNet</p>",
      "rawMarkdown": "Thank you for sharing. I am gonna try to implement it with DenseNet",
      "votes": 1
    },
    {
      "id": 870850,
      "postDate": "2020-06-02T01:39:26.987Z",
      "content": "<p>👍 👍 </p>",
      "rawMarkdown": "👍 👍 ",
      "votes": 1
    },
    {
      "id": 867939,
      "postDate": "2020-05-30T18:42:00.300Z",
      "content": "<p>Great work 👍 </p>",
      "rawMarkdown": "Great work 👍 ",
      "votes": 1
    },
    {
      "id": 876624,
      "postDate": "2020-06-06T20:58:20.303Z",
      "content": "<p>Metadata model\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F666816%2F529bb43d1b229861a9865d8d1915a997%2FCapture%20decran%202020-06-06%20a%2010.57.16%20PM.png?generation=1591477062025054&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Metadata model\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F666816%2F529bb43d1b229861a9865d8d1915a997%2FCapture%20decran%202020-06-06%20a%2010.57.16%20PM.png?generation=1591477062025054&amp;alt=media)\n",
      "votes": 2
    },
    {
      "id": 873547,
      "postDate": "2020-06-04T08:53:29.980Z",
      "content": "<p>Great work!!!!👍 </p>",
      "rawMarkdown": "Great work!!!!👍 ",
      "votes": 2
    },
    {
      "id": 906916,
      "postDate": "2020-06-29T15:56:01.877Z",
      "content": "<p>Guy, can anybody shear actual code? I don't understand this Github stuff?</p>",
      "rawMarkdown": "Guy, can anybody shear actual code? I don't understand this Github stuff?",
      "votes": -5,
      "replies": [
        {
          "id": 909676,
          "postDate": "2020-06-30T18:25:29.473Z",
          "content": "<p>when you click on link on github you will see the actual code, you can manually download each py file or all together</p>",
          "rawMarkdown": "when you click on link on github you will see the actual code, you can manually download each py file or all together",
          "votes": 1
        }
      ]
    },
    {
      "id": 870952,
      "postDate": "2020-06-02T04:03:20.693Z",
      "content": "<p>Can someone please explain to me what is the meta-data that they are referring to ? </p>",
      "rawMarkdown": "Can someone please explain to me what is the meta-data that they are referring to ? \n",
      "replies": [
        {
          "id": 871004,
          "postDate": "2020-06-02T04:56:33.033Z",
          "content": "<p>Meta data is the tabular data <code>gender</code>, <code>age</code>, etc that accompanies the image data. You can use this information in addition to the images. More discussion <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155251\">here</a> and notebook <a href=\"https://www.kaggle.com/cdeotte/image-and-tabular-data-0-915\">here</a></p>",
          "rawMarkdown": "Meta data is the tabular data `gender`, `age`, etc that accompanies the image data. You can use this information in addition to the images. More discussion [here][1] and notebook [here][2]\n\n[1]: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155251\n[2]: https://www.kaggle.com/cdeotte/image-and-tabular-data-0-915",
          "votes": 9
        },
        {
          "id": 871579,
          "postDate": "2020-06-02T13:35:53.887Z",
          "content": "<p>Thank you very much, I'll have a look !</p>",
          "rawMarkdown": "Thank you very much, I'll have a look !"
        }
      ]
    },
    {
      "id": 969477,
      "postDate": "2020-08-13T17:54:30.397Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    },
    {
      "id": 891909,
      "postDate": "2020-06-18T14:25:33.580Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    },
    {
      "id": 953471,
      "postDate": "2020-07-31T20:11:40.983Z",
      "content": "<p>Very well done, thanks for sharing!</p>",
      "rawMarkdown": "Very well done, thanks for sharing!",
      "votes": 1
    },
    {
      "id": 952720,
      "postDate": "2020-07-31T06:59:27.323Z",
      "content": "<p>Good writing! Thanks for sharing :)</p>",
      "rawMarkdown": "Good writing! Thanks for sharing :)",
      "votes": 1
    },
    {
      "id": 952716,
      "postDate": "2020-07-31T06:57:08.977Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!",
      "votes": 1
    },
    {
      "id": 943022,
      "postDate": "2020-07-24T05:24:45.317Z",
      "content": "<p>Thanks for sharing</p>",
      "rawMarkdown": "Thanks for sharing",
      "votes": 1
    },
    {
      "id": 935317,
      "postDate": "2020-07-19T09:16:01.520Z",
      "content": "<p>Thanks. It will help.</p>",
      "rawMarkdown": "Thanks. It will help.",
      "votes": 1
    },
    {
      "id": 916824,
      "postDate": "2020-07-06T04:16:45.980Z",
      "content": "<p>Good to know. Thanks for sharing!</p>",
      "rawMarkdown": "Good to know. Thanks for sharing!",
      "votes": 1
    },
    {
      "id": 874988,
      "postDate": "2020-06-05T12:32:43.107Z",
      "content": "<p>Thanks for sharing! </p>",
      "rawMarkdown": "Thanks for sharing! ",
      "votes": 1
    },
    {
      "id": 866666,
      "postDate": "2020-05-29T15:27:26.130Z",
      "content": "<p>Nice work. Thanks for sharing :)</p>",
      "rawMarkdown": "Nice work. Thanks for sharing :)\n",
      "votes": 1
    },
    {
      "id": 866428,
      "postDate": "2020-05-29T11:52:07.860Z",
      "content": "<p>Great, thanks for sharing! :)</p>",
      "rawMarkdown": "Great, thanks for sharing! :)",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 969631,
      "author_name": "Harsh Kothari21",
      "author_url": "",
      "post_date": "2020-08-13T20:10:58.090000",
      "content": "<p>Nice work!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 963672,
      "author_name": "Amritvir Singh",
      "author_url": "",
      "post_date": "2020-08-09T07:46:11.190000",
      "content": "<p>Thanks for sharing! helped me learn a lot form the ay you picked up insights from the data.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 893542,
      "author_name": "Aditya Baurai",
      "author_url": "",
      "post_date": "2020-06-19T17:51:34.467000",
      "content": "<p>Thank you for sharing. I am gonna try to implement it with DenseNet</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 870850,
      "author_name": "David Wainiqolo",
      "author_url": "",
      "post_date": "2020-06-02T01:39:26.987000",
      "content": "<p>👍 👍 </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 867939,
      "author_name": "Somesh Sharma",
      "author_url": "",
      "post_date": "2020-05-30T18:42:00.300000",
      "content": "<p>Great work 👍 </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 876624,
      "author_name": "Diego Gomez",
      "author_url": "",
      "post_date": "2020-06-06T20:58:20.303000",
      "content": "<p>Metadata model\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F666816%2F529bb43d1b229861a9865d8d1915a997%2FCapture%20decran%202020-06-06%20a%2010.57.16%20PM.png?generation=1591477062025054&amp;alt=media\" alt=\"\"></p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 873547,
      "author_name": "Vaidic Jain",
      "author_url": "",
      "post_date": "2020-06-04T08:53:29.980000",
      "content": "<p>Great work!!!!👍 </p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 906916,
      "author_name": "Serge",
      "author_url": "",
      "post_date": "2020-06-29T15:56:01.877000",
      "content": "<p>Guy, can anybody shear actual code? I don't understand this Github stuff?</p>",
      "votes": -5,
      "replies": [
        {
          "id": 909676,
          "author_name": "Jacek Poplawski",
          "author_url": "",
          "post_date": "2020-06-30T18:25:29.473000",
          "content": "<p>when you click on link on github you will see the actual code, you can manually download each py file or all together</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 870952,
      "author_name": "Naman Makkar",
      "author_url": "",
      "post_date": "2020-06-02T04:03:20.693000",
      "content": "<p>Can someone please explain to me what is the meta-data that they are referring to ? </p>",
      "votes": 0,
      "replies": [
        {
          "id": 871004,
          "author_name": "Chris Deotte",
          "author_url": "",
          "post_date": "2020-06-02T04:56:33.033000",
          "content": "<p>Meta data is the tabular data <code>gender</code>, <code>age</code>, etc that accompanies the image data. You can use this information in addition to the images. More discussion <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155251\">here</a> and notebook <a href=\"https://www.kaggle.com/cdeotte/image-and-tabular-data-0-915\">here</a></p>",
          "votes": 9,
          "replies": []
        },
        {
          "id": 871579,
          "author_name": "Naman Makkar",
          "author_url": "",
          "post_date": "2020-06-02T13:35:53.887000",
          "content": "<p>Thank you very much, I'll have a look !</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 969477,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-08-13T17:54:30.397000",
      "content": "",
      "votes": 1,
      "replies": []
    },
    {
      "id": 891909,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-06-18T14:25:33.580000",
      "content": "",
      "votes": 1,
      "replies": []
    },
    {
      "id": 953471,
      "author_name": "Alin Cijov",
      "author_url": "",
      "post_date": "2020-07-31T20:11:40.983000",
      "content": "<p>Very well done, thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 952720,
      "author_name": "Salman Ibne Eunus",
      "author_url": "",
      "post_date": "2020-07-31T06:59:27.323000",
      "content": "<p>Good writing! Thanks for sharing :)</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 952716,
      "author_name": "Salman Ibne Eunus",
      "author_url": "",
      "post_date": "2020-07-31T06:57:08.977000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 943022,
      "author_name": "FC",
      "author_url": "",
      "post_date": "2020-07-24T05:24:45.317000",
      "content": "<p>Thanks for sharing</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 935317,
      "author_name": "Khan Fashee Monowar (Sawrup)",
      "author_url": "",
      "post_date": "2020-07-19T09:16:01.520000",
      "content": "<p>Thanks. It will help.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 916824,
      "author_name": "Helen",
      "author_url": "",
      "post_date": "2020-07-06T04:16:45.980000",
      "content": "<p>Good to know. Thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 874988,
      "author_name": "Codefupanda",
      "author_url": "",
      "post_date": "2020-06-05T12:32:43.107000",
      "content": "<p>Thanks for sharing! </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 866666,
      "author_name": "Navin",
      "author_url": "",
      "post_date": "2020-05-29T15:27:26.130000",
      "content": "<p>Nice work. Thanks for sharing :)</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 866428,
      "author_name": "Gajendra Saraswat",
      "author_url": "",
      "post_date": "2020-05-29T11:52:07.860000",
      "content": "<p>Great, thanks for sharing! :)</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "866410": "ISIC 2019 challenge 1st place solution [paper](https://doi.org/10.1016/j.mex.2020.100864) and [PyTorch code](https://github.com/ngessert/isic2019)\n\nKey insights:\n- preprocessing: cropping + Shades of Gray color constancy\n-  EfficientNets ([EfficientNet-PyTorch](https://github.com/lukemelas/EfficientNet-PyTorch)) + SENet154\n- extensive augs + CutOut\n- ensembling that maximizes the mean sensitivity\n- MLP for meta-data\n\n![network architecture](https://ars.els-cdn.com/content/image/1-s2.0-S2215016120300832-gr2.jpg)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F66669%2Fb08311fcc713bc4796fe24ab28872b93%2FScreen%20Shot%202020-05-29%20at%201.25.47%20PM.png?generation=1590750836068470&amp;alt=media)\n\n\n[ISIC 2019 leaderboard](https://challenge2019.isic-archive.com/leaderboard.html) contains links to papers describing various other approaches. E.g. 3rd place built  an ensemble of leave out classifiers.\n",
    "969631": "Nice work!",
    "963672": "Thanks for sharing! helped me learn a lot form the ay you picked up insights from the data.\n",
    "893542": "Thank you for sharing. I am gonna try to implement it with DenseNet",
    "870850": "👍 👍 ",
    "867939": "Great work 👍 ",
    "876624": "Metadata model\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F666816%2F529bb43d1b229861a9865d8d1915a997%2FCapture%20decran%202020-06-06%20a%2010.57.16%20PM.png?generation=1591477062025054&amp;alt=media)\n",
    "873547": "Great work!!!!👍 ",
    "906916": "Guy, can anybody shear actual code? I don't understand this Github stuff?",
    "870952": "Can someone please explain to me what is the meta-data that they are referring to ? \n",
    "969477": "",
    "891909": "",
    "953471": "Very well done, thanks for sharing!",
    "952720": "Good writing! Thanks for sharing :)",
    "952716": "Thanks for sharing!",
    "943022": "Thanks for sharing",
    "935317": "Thanks. It will help.",
    "916824": "Good to know. Thanks for sharing!",
    "874988": "Thanks for sharing! ",
    "866666": "Nice work. Thanks for sharing :)\n",
    "866428": "Great, thanks for sharing! :)"
  }
}