{
  "id": 316782,
  "title": "Metric learning is forgotten forever?",
  "url": "/competitions/happy-whale-and-dolphin/discussion/316782",
  "author_name": "",
  "post_date": "2022-04-03T20:32:33.244223900Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>A lot of notebooks with ArcFace and no one with triple loss? Is it no longer used today?</p>",
  "messages": [
    {
      "id": "1744304",
      "postDate": "04/03/2022 20:32:33",
      "content": "<p>A lot of notebooks with ArcFace and no one with triple loss? Is it no longer used today?</p>",
      "rawMarkdown": "A lot of notebooks with ArcFace and no one with triple loss? Is it no longer used today?",
      "votes": null
    },
    {
      "id": "1744723",
      "postDate": "04/04/2022 08:37:53",
      "content": "<p>The main problem of triplet - hard sample mining. With unlimited time triplet may be better then margin-based softmax, but usually we don't have unlimited time and currently we don't have some optimal algorithms for hard sample mining. If we fix time - ArcFace is usually much better.</p>",
      "rawMarkdown": "The main problem of triplet - hard sample mining. With unlimited time triplet may be better then margin-based softmax, but usually we don't have unlimited time and currently we don't have some optimal algorithms for hard sample mining. If we fix time - ArcFace is usually much better.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1744723,
      "author_name": "kwentar",
      "author_url": "",
      "post_date": "04/04/2022 08:37:53",
      "content": "<p>The main problem of triplet - hard sample mining. With unlimited time triplet may be better then margin-based softmax, but usually we don't have unlimited time and currently we don't have some optimal algorithms for hard sample mining. If we fix time - ArcFace is usually much better.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1744304": "A lot of notebooks with ArcFace and no one with triple loss? Is it no longer used today?",
    "1744723": "The main problem of triplet - hard sample mining. With unlimited time triplet may be better then margin-based softmax, but usually we don't have unlimited time and currently we don't have some optimal algorithms for hard sample mining. If we fix time - ArcFace is usually much better."
  },
  "source": "meta"
}