{
  "id": 425360,
  "title": "19th Place Solution - Simple SP+SG via HLoc",
  "url": "/competitions/image-matching-challenge-2023/writeups/to-be-worst-19th-place-solution-simple-sp-sg-via-h",
  "author_name": "",
  "post_date": "2023-07-18T11:41:29.205465900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Our solution is based on the baseline. In addition, we used the famous open sourced code named “Hierarchical-Localization-master\" (<a href=\"https://github.com/cvg/Hierarchical-Localization\" target=\"_blank\">https://github.com/cvg/Hierarchical-Localization</a>) to implement a famous image matching pipeline named \"SuperPoint + SuperGlue\" without fine tuning but only pretrained weights. So we spent no time on training. We changed parameters such as nms_radius, max_keypoints, resize_max and sinkhorn_iterations to achieve our best leaderboard score.</p>",
  "messages": [
    {
      "id": "2349400",
      "postDate": "07/18/2023 11:41:29",
      "content": "<p>Our solution is based on the baseline. In addition, we used the famous open sourced code named “Hierarchical-Localization-master\" (<a href=\"https://github.com/cvg/Hierarchical-Localization\" target=\"_blank\">https://github.com/cvg/Hierarchical-Localization</a>) to implement a famous image matching pipeline named \"SuperPoint + SuperGlue\" without fine tuning but only pretrained weights. So we spent no time on training. We changed parameters such as nms_radius, max_keypoints, resize_max and sinkhorn_iterations to achieve our best leaderboard score.</p>",
      "rawMarkdown": "Our solution is based on the baseline. In addition, we used the famous open sourced code named “Hierarchical-Localization-master\" (https://github.com/cvg/Hierarchical-Localization) to implement a famous image matching pipeline named \"SuperPoint + SuperGlue\" without fine tuning but only pretrained weights. So we spent no time on training. We changed parameters such as nms_radius, max_keypoints, resize_max and sinkhorn_iterations to achieve our best leaderboard score.",
      "votes": null
    },
    {
      "id": "2349401",
      "postDate": "07/18/2023 11:42:03",
      "content": "<p>The github link of HLoc: <a href=\"https://github.com/cvg/Hierarchical-Localization\" target=\"_blank\">https://github.com/cvg/Hierarchical-Localization</a></p>",
      "rawMarkdown": "The github link of HLoc: https://github.com/cvg/Hierarchical-Localization",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2349401,
      "author_name": "zhongwenhao",
      "author_url": "",
      "post_date": "07/18/2023 11:42:03",
      "content": "<p>The github link of HLoc: <a href=\"https://github.com/cvg/Hierarchical-Localization\" target=\"_blank\">https://github.com/cvg/Hierarchical-Localization</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2349400": "Our solution is based on the baseline. In addition, we used the famous open sourced code named “Hierarchical-Localization-master\" (https://github.com/cvg/Hierarchical-Localization) to implement a famous image matching pipeline named \"SuperPoint + SuperGlue\" without fine tuning but only pretrained weights. So we spent no time on training. We changed parameters such as nms_radius, max_keypoints, resize_max and sinkhorn_iterations to achieve our best leaderboard score.",
    "2349401": "The github link of HLoc: https://github.com/cvg/Hierarchical-Localization"
  },
  "source": "meta"
}