{
  "id": 163654,
  "title": "Embedding?",
  "url": "/competitions/landmark-retrieval-2020/discussion/163654",
  "author_name": "",
  "post_date": "2020-07-02T22:40:30.697583100Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>So as far as i have understand the objective is to upload a model that has an embedding? If the answer is yes i dont understand why we need the landmarks, indexes etc...</p>",
  "messages": [
    {
      "id": "913030",
      "postDate": "07/02/2020 22:40:30",
      "content": "<p>So as far as i have understand the objective is to upload a model that has an embedding? If the answer is yes i dont understand why we need the landmarks, indexes etc...</p>",
      "rawMarkdown": "So as far as i have understand the objective is to upload a model that has an embedding? If the answer is yes i dont understand why we need the landmarks, indexes etc...",
      "votes": null
    },
    {
      "id": "913640",
      "postDate": "07/03/2020 10:27:20",
      "content": "<p>Hi <a href=\"/enric1296\">@enric1296</a>,</p>\n\n<p>This is my take on this, I hope it matches the organizer point of view.</p>\n\n<p>The embedding of each image will be used by a k-NN algorithm to create clusters of similar images. To these clusters will then be appointed a class. This k-NN algorithm will then be able to measure classification performance by checking how many images are appointed to incorrect clusters.</p>\n\n<p>This task is inherently linked to <a href=\"https://en.wikipedia.org/wiki/Similarity_learning\">Similarity Learning</a>, I think, and you should definitely check this out.</p>",
      "rawMarkdown": "Hi @enric1296,\n\nThis is my take on this, I hope it matches the organizer point of view.\n\nThe embedding of each image will be used by a k-NN algorithm to create clusters of similar images. To these clusters will then be appointed a class. This k-NN algorithm will then be able to measure classification performance by checking how many images are appointed to incorrect clusters.\n\nThis task is inherently linked to [Similarity Learning](https://en.wikipedia.org/wiki/Similarity_learning), I think, and you should definitely check this out.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 913640,
      "author_name": "dimartinot",
      "author_url": "",
      "post_date": "07/03/2020 10:27:20",
      "content": "<p>Hi <a href=\"/enric1296\">@enric1296</a>,</p>\n\n<p>This is my take on this, I hope it matches the organizer point of view.</p>\n\n<p>The embedding of each image will be used by a k-NN algorithm to create clusters of similar images. To these clusters will then be appointed a class. This k-NN algorithm will then be able to measure classification performance by checking how many images are appointed to incorrect clusters.</p>\n\n<p>This task is inherently linked to <a href=\"https://en.wikipedia.org/wiki/Similarity_learning\">Similarity Learning</a>, I think, and you should definitely check this out.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "913030": "So as far as i have understand the objective is to upload a model that has an embedding? If the answer is yes i dont understand why we need the landmarks, indexes etc...",
    "913640": "Hi @enric1296,\n\nThis is my take on this, I hope it matches the organizer point of view.\n\nThe embedding of each image will be used by a k-NN algorithm to create clusters of similar images. To these clusters will then be appointed a class. This k-NN algorithm will then be able to measure classification performance by checking how many images are appointed to incorrect clusters.\n\nThis task is inherently linked to [Similarity Learning](https://en.wikipedia.org/wiki/Similarity_learning), I think, and you should definitely check this out."
  },
  "source": "meta"
}