{
  "id": 314137,
  "title": "Is arcface the best solution?",
  "url": "/competitions/happy-whale-and-dolphin/discussion/314137",
  "author_name": "",
  "post_date": "2022-03-21T07:11:56.982042900Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Is arcface the best solution?Did you try to use other losses?</p>",
  "messages": [
    {
      "id": "1730356",
      "postDate": "03/21/2022 07:11:56",
      "content": "<p>Is arcface the best solution?Did you try to use other losses?</p>",
      "rawMarkdown": "Is arcface the best solution?Did you try to use other losses?",
      "votes": null
    },
    {
      "id": "1733772",
      "postDate": "03/24/2022 15:56:28",
      "content": "<p>I have tried two so far - ArcFace and CosFace and I have seen much worse (CV &amp; LB) in CosFace compared to ArcFace. For next experiments I will be trying out Focal &amp; Spherical Loss.</p>",
      "rawMarkdown": "I have tried two so far - ArcFace and CosFace and I have seen much worse (CV & LB) in CosFace compared to ArcFace. For next experiments I will be trying out Focal & Spherical Loss.",
      "votes": null
    },
    {
      "id": "1733780",
      "postDate": "03/24/2022 16:08:42",
      "content": "<p>I've also tried CurricularFace but I still  seen much more worse result too @@ I think evolution of similarity loss function just base on face identity problem so it may not good at other identity task (I dont know @@) </p>",
      "rawMarkdown": "I've also tried CurricularFace but I still  seen much more worse result too @@ I think evolution of similarity loss function just base on face identity problem so it may not good at other identity task (I dont know @@)",
      "votes": null
    },
    {
      "id": "1733929",
      "postDate": "03/24/2022 18:47:01",
      "content": "<p>I tested AdaCos and Arcface<br>\nArcFace was better (around 0.01 in score)</p>",
      "rawMarkdown": "I tested AdaCos and Arcface\nArcFace was better (around 0.01 in score)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1733772,
      "author_name": "thakurudit",
      "author_url": "",
      "post_date": "03/24/2022 15:56:28",
      "content": "<p>I have tried two so far - ArcFace and CosFace and I have seen much worse (CV &amp; LB) in CosFace compared to ArcFace. For next experiments I will be trying out Focal &amp; Spherical Loss.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1733780,
          "author_name": "dqhdqmcttdqx",
          "author_url": "",
          "post_date": "03/24/2022 16:08:42",
          "content": "<p>I've also tried CurricularFace but I still  seen much more worse result too @@ I think evolution of similarity loss function just base on face identity problem so it may not good at other identity task (I dont know @@) </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1733929,
      "author_name": "deepkim",
      "author_url": "",
      "post_date": "03/24/2022 18:47:01",
      "content": "<p>I tested AdaCos and Arcface<br>\nArcFace was better (around 0.01 in score)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1730356": "Is arcface the best solution?Did you try to use other losses?",
    "1733772": "I have tried two so far - ArcFace and CosFace and I have seen much worse (CV & LB) in CosFace compared to ArcFace. For next experiments I will be trying out Focal & Spherical Loss.",
    "1733780": "I've also tried CurricularFace but I still  seen much more worse result too @@ I think evolution of similarity loss function just base on face identity problem so it may not good at other identity task (I dont know @@)",
    "1733929": "I tested AdaCos and Arcface\nArcFace was better (around 0.01 in score)"
  },
  "source": "meta"
}