{
  "id": 135108,
  "title": "Equalization Loss for Long-Tailed Object Recognition",
  "url": "/competitions/bengaliai-cv19/discussion/135108",
  "author_name": "",
  "post_date": "2020-03-12T03:45:56.964090300Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p><img src=\"https://camo.githubusercontent.com/962dd266c2b9783c4ad86503ad66163702f9fe7b/68747470733a2f2f747a747a747a747a747a2e6769746875622e696f2f696d616765732f65716c2d6772616469656e742e6a7067\" alt=\"1\"></p>\n\n<p><a href=\"https://github.com/tztztztztz/eql.detectron2\">Github</a></p>\n\n<p><a href=\"https://128.84.21.199/abs/2003.05176\">arXiv</a></p>\n\n<p>EQL protects the learning for rare categories from being at a disadvantage during the network parameter updating under the long-tailed situation.</p>\n\n<p>While the original idea is for object recognition, EQL may help recognize rare class bengali characters for this competition.</p>",
  "messages": [
    {
      "id": "769596",
      "postDate": "03/12/2020 03:45:56",
      "content": "<p><img src=\"https://camo.githubusercontent.com/962dd266c2b9783c4ad86503ad66163702f9fe7b/68747470733a2f2f747a747a747a747a747a2e6769746875622e696f2f696d616765732f65716c2d6772616469656e742e6a7067\" alt=\"1\"></p>\n\n<p><a href=\"https://github.com/tztztztztz/eql.detectron2\">Github</a></p>\n\n<p><a href=\"https://128.84.21.199/abs/2003.05176\">arXiv</a></p>\n\n<p>EQL protects the learning for rare categories from being at a disadvantage during the network parameter updating under the long-tailed situation.</p>\n\n<p>While the original idea is for object recognition, EQL may help recognize rare class bengali characters for this competition.</p>",
      "rawMarkdown": "![1](https://camo.githubusercontent.com/962dd266c2b9783c4ad86503ad66163702f9fe7b/68747470733a2f2f747a747a747a747a747a2e6769746875622e696f2f696d616765732f65716c2d6772616469656e742e6a7067)\n\n[Github](https://github.com/tztztztztz/eql.detectron2)\n\n[arXiv](https://128.84.21.199/abs/2003.05176)\n\nEQL protects the learning for rare categories from being at a disadvantage during the network parameter updating under the long-tailed situation.\n\nWhile the original idea is for object recognition, EQL may help recognize rare class bengali characters for this competition.",
      "votes": null
    },
    {
      "id": "769718",
      "postDate": "03/12/2020 07:15:32",
      "content": "<p>thanks for your kindly sharing</p>",
      "rawMarkdown": "thanks for your kindly sharing",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 769718,
      "author_name": "karlyukang",
      "author_url": "",
      "post_date": "03/12/2020 07:15:32",
      "content": "<p>thanks for your kindly sharing</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "769596": "![1](https://camo.githubusercontent.com/962dd266c2b9783c4ad86503ad66163702f9fe7b/68747470733a2f2f747a747a747a747a747a2e6769746875622e696f2f696d616765732f65716c2d6772616469656e742e6a7067)\n\n[Github](https://github.com/tztztztztz/eql.detectron2)\n\n[arXiv](https://128.84.21.199/abs/2003.05176)\n\nEQL protects the learning for rare categories from being at a disadvantage during the network parameter updating under the long-tailed situation.\n\nWhile the original idea is for object recognition, EQL may help recognize rare class bengali characters for this competition.",
    "769718": "thanks for your kindly sharing"
  },
  "source": "meta"
}