{
  "id": 402497,
  "title": "Frame of reference for each scene",
  "url": "/competitions/image-matching-challenge-2023/discussion/402497",
  "author_name": "",
  "post_date": "2023-04-18T16:13:43.986395400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi, hope you guys are fine and well.</p>\n<p>I have a confusion regarding the statement <br>\n<strong>Each camera pose is parameterized with a rotation matrix R and a translation vector T from an arbitrary frame of reference</strong><br>\nDoes this means that we have different frame of reference for each image in a single scene (which does not make sense to me because the mAA is computed based on the comparison between pairs of images from the same scene that implies common frame of reference) or is it the same for each images in a single scene but different across multiple scene? </p>\n<p>Thank you. </p>",
  "messages": [
    {
      "id": "2226065",
      "postDate": "04/18/2023 16:13:43",
      "content": "<p>Hi, hope you guys are fine and well.</p>\n<p>I have a confusion regarding the statement <br>\n<strong>Each camera pose is parameterized with a rotation matrix R and a translation vector T from an arbitrary frame of reference</strong><br>\nDoes this means that we have different frame of reference for each image in a single scene (which does not make sense to me because the mAA is computed based on the comparison between pairs of images from the same scene that implies common frame of reference) or is it the same for each images in a single scene but different across multiple scene? </p>\n<p>Thank you. </p>",
      "rawMarkdown": "Hi, hope you guys are fine and well.\n\nI have a confusion regarding the statement \n**Each camera pose is parameterized with a rotation matrix R and a translation vector T from an arbitrary frame of reference**\nDoes this means that we have different frame of reference for each image in a single scene (which does not make sense to me because the mAA is computed based on the comparison between pairs of images from the same scene that implies common frame of reference) or is it the same for each images in a single scene but different across multiple scene? \n\nThank you.",
      "votes": null
    },
    {
      "id": "2226076",
      "postDate": "04/18/2023 16:23:35",
      "content": "<p>Sorry, could you please specify where this statement is reported? It's difficult without a context to say something…</p>",
      "rawMarkdown": "Sorry, could you please specify where this statement is reported? It's difficult without a context to say something...",
      "votes": null
    },
    {
      "id": "2226217",
      "postDate": "04/18/2023 18:48:42",
      "content": "<p>Its located in the <strong>Overview</strong> -&gt; <strong>Evaluation</strong> -&gt; <strong>Evaluation Metric</strong> section</p>",
      "rawMarkdown": "Its located in the **Overview** -> **Evaluation** -> **Evaluation Metric** section",
      "votes": null
    },
    {
      "id": "2226239",
      "postDate": "04/18/2023 19:10:10",
      "content": "<p>The 3D reference system is up to a similarity transformation (scale, rotation and translation). In practice, COLMAP camera poses (and 3D point cloud) of the GT will differ from those submitted by a similarity transformation. Since the evaluation is pairwise so that the all the RELATIVE pairwise transformations between cameras are compared, NOT the ABSOLUTE poses, there is no problem. Overall, up to the similarity transform, LOCAL relative poses are GLOBALLY CONSISTENT.</p>\n<p>You can check the evaluation code provided (<a href=\"https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation\" target=\"_blank\">https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation</a>) for more insight!</p>",
      "rawMarkdown": "The 3D reference system is up to a similarity transformation (scale, rotation and translation). In practice, COLMAP camera poses (and 3D point cloud) of the GT will differ from those submitted by a similarity transformation. Since the evaluation is pairwise so that the all the RELATIVE pairwise transformations between cameras are compared, NOT the ABSOLUTE poses, there is no problem. Overall, up to the similarity transform, LOCAL relative poses are GLOBALLY CONSISTENT.\n\nYou can check the evaluation code provided (https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation) for more insight!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2226076,
      "author_name": "fabiobellavia",
      "author_url": "",
      "post_date": "04/18/2023 16:23:35",
      "content": "<p>Sorry, could you please specify where this statement is reported? It's difficult without a context to say something…</p>",
      "votes": null,
      "replies": [
        {
          "id": 2226217,
          "author_name": "mohammadasimbluemoon",
          "author_url": "",
          "post_date": "04/18/2023 18:48:42",
          "content": "<p>Its located in the <strong>Overview</strong> -&gt; <strong>Evaluation</strong> -&gt; <strong>Evaluation Metric</strong> section</p>",
          "votes": null,
          "replies": [
            {
              "id": 2226239,
              "author_name": "fabiobellavia",
              "author_url": "",
              "post_date": "04/18/2023 19:10:10",
              "content": "<p>The 3D reference system is up to a similarity transformation (scale, rotation and translation). In practice, COLMAP camera poses (and 3D point cloud) of the GT will differ from those submitted by a similarity transformation. Since the evaluation is pairwise so that the all the RELATIVE pairwise transformations between cameras are compared, NOT the ABSOLUTE poses, there is no problem. Overall, up to the similarity transform, LOCAL relative poses are GLOBALLY CONSISTENT.</p>\n<p>You can check the evaluation code provided (<a href=\"https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation\" target=\"_blank\">https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation</a>) for more insight!</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2226065": "Hi, hope you guys are fine and well.\n\nI have a confusion regarding the statement \n**Each camera pose is parameterized with a rotation matrix R and a translation vector T from an arbitrary frame of reference**\nDoes this means that we have different frame of reference for each image in a single scene (which does not make sense to me because the mAA is computed based on the comparison between pairs of images from the same scene that implies common frame of reference) or is it the same for each images in a single scene but different across multiple scene? \n\nThank you.",
    "2226076": "Sorry, could you please specify where this statement is reported? It's difficult without a context to say something...",
    "2226217": "Its located in the **Overview** -> **Evaluation** -> **Evaluation Metric** section",
    "2226239": "The 3D reference system is up to a similarity transformation (scale, rotation and translation). In practice, COLMAP camera poses (and 3D point cloud) of the GT will differ from those submitted by a similarity transformation. Since the evaluation is pairwise so that the all the RELATIVE pairwise transformations between cameras are compared, NOT the ABSOLUTE poses, there is no problem. Overall, up to the similarity transform, LOCAL relative poses are GLOBALLY CONSISTENT.\n\nYou can check the evaluation code provided (https://www.kaggle.com/code/eduardtrulls/imc2023-evaluation) for more insight!"
  },
  "source": "meta"
}