{
  "id": 234718,
  "title": "how to train using test images : self-supervised tricks that every kaggler should know",
  "url": "/competitions/bms-molecular-translation/discussion/234718",
  "author_name": "",
  "post_date": "2021-04-25T19:32:40.736174700Z",
  "votes": 35,
  "comment_count": 5,
  "views": 0,
  "content": "<p>there are many ways to do this. i will add a few ideas over the next few days. Here is the first one<br>\n<img src=\"https://i.ibb.co/d5K0qzg/Selection-148.png\" alt=\"\"></p>",
  "messages": [
    {
      "id": "1284316",
      "postDate": "04/25/2021 19:32:40",
      "content": "<p>there are many ways to do this. i will add a few ideas over the next few days. Here is the first one<br>\n<img src=\"https://i.ibb.co/d5K0qzg/Selection-148.png\" alt=\"\"></p>",
      "rawMarkdown": "there are many ways to do this. i will add a few ideas over the next few days. Here is the first one\n![](https://i.ibb.co/d5K0qzg/Selection-148.png)",
      "votes": null
    },
    {
      "id": "1284325",
      "postDate": "04/25/2021 19:37:45",
      "content": "<p><img src=\"https://i.ibb.co/GC1MCJp/Selection-149.png\" alt=\"\"></p>",
      "rawMarkdown": "![](https://i.ibb.co/GC1MCJp/Selection-149.png)",
      "votes": null
    },
    {
      "id": "1284344",
      "postDate": "04/25/2021 19:58:39",
      "content": "<p><img src=\"https://i.ibb.co/7nQ1STv/Selection-151.png\" alt=\"\"></p>\n<p>typo error: remove or add items</p>\n<hr>\n<p>in theory you can mix train images and treat them as unlabelled. in this way maybe you can crate more large images (long sequence)???</p>",
      "rawMarkdown": "![](https://i.ibb.co/7nQ1STv/Selection-151.png)\n\ntypo error: remove or add items\n\n---\n\nin theory you can mix train images and treat them as unlabelled. in this way maybe you can crate more large images (long sequence)???",
      "votes": null
    },
    {
      "id": "1284345",
      "postDate": "04/25/2021 19:59:16",
      "content": "<p><img src=\"https://i.ibb.co/02pBp7q/Selection-150.png\" alt=\"\"></p>",
      "rawMarkdown": "![](https://i.ibb.co/02pBp7q/Selection-150.png)",
      "votes": null
    },
    {
      "id": "1284353",
      "postDate": "04/25/2021 20:05:59",
      "content": "<p><img src=\"https://i.ibb.co/LvNwMn6/Selection-152.png\" alt=\"\"></p>",
      "rawMarkdown": "![](https://i.ibb.co/LvNwMn6/Selection-152.png)",
      "votes": null
    },
    {
      "id": "1299833",
      "postDate": "05/10/2021 04:42:31",
      "content": "<p>bidirectional transformer ensemble:<br>\n<a href=\"https://arxiv.org/pdf/2105.02412.pdf\" target=\"_blank\">https://arxiv.org/pdf/2105.02412.pdf</a></p>\n<p>Handwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer</p>\n<p><a href=\"https://github.com/Green-Wood/BTTR\" target=\"_blank\">https://github.com/Green-Wood/BTTR</a></p>",
      "rawMarkdown": "bidirectional transformer ensemble:\nhttps://arxiv.org/pdf/2105.02412.pdf\n\nHandwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer\n\nhttps://github.com/Green-Wood/BTTR",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1284325,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "04/25/2021 19:37:45",
      "content": "<p><img src=\"https://i.ibb.co/GC1MCJp/Selection-149.png\" alt=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1284344,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "04/25/2021 19:58:39",
      "content": "<p><img src=\"https://i.ibb.co/7nQ1STv/Selection-151.png\" alt=\"\"></p>\n<p>typo error: remove or add items</p>\n<hr>\n<p>in theory you can mix train images and treat them as unlabelled. in this way maybe you can crate more large images (long sequence)???</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1284345,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "04/25/2021 19:59:16",
      "content": "<p><img src=\"https://i.ibb.co/02pBp7q/Selection-150.png\" alt=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1284353,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "04/25/2021 20:05:59",
      "content": "<p><img src=\"https://i.ibb.co/LvNwMn6/Selection-152.png\" alt=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1299833,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "05/10/2021 04:42:31",
      "content": "<p>bidirectional transformer ensemble:<br>\n<a href=\"https://arxiv.org/pdf/2105.02412.pdf\" target=\"_blank\">https://arxiv.org/pdf/2105.02412.pdf</a></p>\n<p>Handwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer</p>\n<p><a href=\"https://github.com/Green-Wood/BTTR\" target=\"_blank\">https://github.com/Green-Wood/BTTR</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1284316": "there are many ways to do this. i will add a few ideas over the next few days. Here is the first one\n![](https://i.ibb.co/d5K0qzg/Selection-148.png)",
    "1284325": "![](https://i.ibb.co/GC1MCJp/Selection-149.png)",
    "1284344": "![](https://i.ibb.co/7nQ1STv/Selection-151.png)\n\ntypo error: remove or add items\n\n---\n\nin theory you can mix train images and treat them as unlabelled. in this way maybe you can crate more large images (long sequence)???",
    "1284345": "![](https://i.ibb.co/02pBp7q/Selection-150.png)",
    "1284353": "![](https://i.ibb.co/LvNwMn6/Selection-152.png)",
    "1299833": "bidirectional transformer ensemble:\nhttps://arxiv.org/pdf/2105.02412.pdf\n\nHandwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer\n\nhttps://github.com/Green-Wood/BTTR"
  },
  "source": "meta"
}