{
  "id": 307547,
  "title": "Sub-Center ArcFace w/ Dynamic Margins",
  "url": "/competitions/happy-whale-and-dolphin/discussion/307547",
  "author_name": "",
  "post_date": "2022-02-14T17:48:07.180411900Z",
  "votes": 18,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi all, I've been reviewing papers, previous competition solutions and state-of-the-art in this field.</p>\n<p>It seems pretty clear from the Google Landmark Retrieval 2021 Solutions and from other Contrastive Learning challenges that using Sub-Center ArcFace w/ Dynamic Margins (I BELIEVE the credit goes here for the Dynamic Margins -&gt; <a href=\"https://arxiv.org/pdf/2010.05350.pdf\" target=\"_blank\">https://arxiv.org/pdf/2010.05350.pdf</a> and credit for the Sub-Center part goes to the ArcFace original authors -&gt; <a href=\"https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf\" target=\"_blank\">https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf</a>)</p>\n<p>I can find many implementations of these techniques in Pytorch, but none in Tensorflow. I did find this repo -&gt; <a href=\"https://github.com/leondgarse/Keras_insightface\" target=\"_blank\">https://github.com/leondgarse/Keras_insightface</a>, which, according to its readme…</p>\n<blockquote>\n  <p>[…] is the keras implementation of <a href=\"https://github.com/deepinsight/insightface\" target=\"_blank\">deepinsight/insightface</a> […]</p>\n</blockquote>\n<p>Is anyone aware of other potential implementations in TensorFlow? I may take it up if I can't find anything and use the Pytorch implementations and the leondgarse repo as reference.</p>\n<p>Thanks in advance!</p>",
  "messages": [
    {
      "id": "1690111",
      "postDate": "02/14/2022 17:48:07",
      "content": "<p>Hi all, I've been reviewing papers, previous competition solutions and state-of-the-art in this field.</p>\n<p>It seems pretty clear from the Google Landmark Retrieval 2021 Solutions and from other Contrastive Learning challenges that using Sub-Center ArcFace w/ Dynamic Margins (I BELIEVE the credit goes here for the Dynamic Margins -&gt; <a href=\"https://arxiv.org/pdf/2010.05350.pdf\" target=\"_blank\">https://arxiv.org/pdf/2010.05350.pdf</a> and credit for the Sub-Center part goes to the ArcFace original authors -&gt; <a href=\"https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf\" target=\"_blank\">https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf</a>)</p>\n<p>I can find many implementations of these techniques in Pytorch, but none in Tensorflow. I did find this repo -&gt; <a href=\"https://github.com/leondgarse/Keras_insightface\" target=\"_blank\">https://github.com/leondgarse/Keras_insightface</a>, which, according to its readme…</p>\n<blockquote>\n  <p>[…] is the keras implementation of <a href=\"https://github.com/deepinsight/insightface\" target=\"_blank\">deepinsight/insightface</a> […]</p>\n</blockquote>\n<p>Is anyone aware of other potential implementations in TensorFlow? I may take it up if I can't find anything and use the Pytorch implementations and the leondgarse repo as reference.</p>\n<p>Thanks in advance!</p>",
      "rawMarkdown": "Hi all, I've been reviewing papers, previous competition solutions and state-of-the-art in this field.\n\nIt seems pretty clear from the Google Landmark Retrieval 2021 Solutions and from other Contrastive Learning challenges that using Sub-Center ArcFace w/ Dynamic Margins (I BELIEVE the credit goes here for the Dynamic Margins -> https://arxiv.org/pdf/2010.05350.pdf and credit for the Sub-Center part goes to the ArcFace original authors -> https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf)\n\nI can find many implementations of these techniques in Pytorch, but none in Tensorflow. I did find this repo -> https://github.com/leondgarse/Keras_insightface, which, according to its readme...\n\n> [...] is the keras implementation of [deepinsight/insightface](https://github.com/deepinsight/insightface) [...]\n\nIs anyone aware of other potential implementations in TensorFlow? I may take it up if I can't find anything and use the Pytorch implementations and the leondgarse repo as reference.\n\nThanks in advance!",
      "votes": null
    },
    {
      "id": "1845420",
      "postDate": "07/06/2022 10:14:45",
      "content": "<p>Hey Darien, I am also in need for a tensorflow implementation. Did you find one in the end? If so, I would be very greatful if you could share it with me.  </p>",
      "rawMarkdown": "Hey Darien, I am also in need for a tensorflow implementation. Did you find one in the end? If so, I would be very greatful if you could share it with me.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1845420,
      "author_name": "maximvlah",
      "author_url": "",
      "post_date": "07/06/2022 10:14:45",
      "content": "<p>Hey Darien, I am also in need for a tensorflow implementation. Did you find one in the end? If so, I would be very greatful if you could share it with me.  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1690111": "Hi all, I've been reviewing papers, previous competition solutions and state-of-the-art in this field.\n\nIt seems pretty clear from the Google Landmark Retrieval 2021 Solutions and from other Contrastive Learning challenges that using Sub-Center ArcFace w/ Dynamic Margins (I BELIEVE the credit goes here for the Dynamic Margins -> https://arxiv.org/pdf/2010.05350.pdf and credit for the Sub-Center part goes to the ArcFace original authors -> https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123560715.pdf)\n\nI can find many implementations of these techniques in Pytorch, but none in Tensorflow. I did find this repo -> https://github.com/leondgarse/Keras_insightface, which, according to its readme...\n\n> [...] is the keras implementation of [deepinsight/insightface](https://github.com/deepinsight/insightface) [...]\n\nIs anyone aware of other potential implementations in TensorFlow? I may take it up if I can't find anything and use the Pytorch implementations and the leondgarse repo as reference.\n\nThanks in advance!",
    "1845420": "Hey Darien, I am also in need for a tensorflow implementation. Did you find one in the end? If so, I would be very greatful if you could share it with me."
  },
  "source": "meta"
}