{
  "id": 501788,
  "title": "The Chamfer distance metric",
  "url": "/competitions/cvpr-metafood-3d-food-reconstruction-challenge/discussion/501788",
  "author_name": "",
  "post_date": "2024-05-10T19:10:26.958248100Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Comparing meshes using Chamfer distance is helpful, but the distance may be significantly higher if the meshes (i.e., the predicted mesh against ground truth mesh) are not aligned due to rotations or translations. From which angle should we present the predicted meshes? From the top, similar to the overhead photos?</p>",
  "messages": [
    {
      "id": "2805859",
      "postDate": "05/10/2024 19:10:26",
      "content": "<p>Comparing meshes using Chamfer distance is helpful, but the distance may be significantly higher if the meshes (i.e., the predicted mesh against ground truth mesh) are not aligned due to rotations or translations. From which angle should we present the predicted meshes? From the top, similar to the overhead photos?</p>",
      "rawMarkdown": "Comparing meshes using Chamfer distance is helpful, but the distance may be significantly higher if the meshes (i.e., the predicted mesh against ground truth mesh) are not aligned due to rotations or translations. From which angle should we present the predicted meshes? From the top, similar to the overhead photos?",
      "votes": null
    },
    {
      "id": "2809756",
      "postDate": "05/13/2024 01:22:00",
      "content": "<p>Once we pass the first phase, we will be providing our evaluation camera parameters for each image. You will be asked to send us your translated object based on our camera parameters. You solution will be compared to our ground truth (collected by a 3D scanner). The rotation and translation of our ground truth object are manually labeled to align with the object in the image. </p>\n<p>For the final evaluation, we will take the minimum chamfer distance produced by three kinds of translation. Each translation will be computed to translate your object to the ground truth object. These translations are (1) no translation, (2) translation by Point-to-point ICP from open3D library, (3) translation by Point-to-plane ICP from open3D library. </p>",
      "rawMarkdown": "Once we pass the first phase, we will be providing our evaluation camera parameters for each image. You will be asked to send us your translated object based on our camera parameters. You solution will be compared to our ground truth (collected by a 3D scanner). The rotation and translation of our ground truth object are manually labeled to align with the object in the image. \n\nFor the final evaluation, we will take the minimum chamfer distance produced by three kinds of translation. Each translation will be computed to translate your object to the ground truth object. These translations are (1) no translation, (2) translation by Point-to-point ICP from open3D library, (3) translation by Point-to-plane ICP from open3D library.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2809756,
      "author_name": "metafoodcvpr",
      "author_url": "",
      "post_date": "05/13/2024 01:22:00",
      "content": "<p>Once we pass the first phase, we will be providing our evaluation camera parameters for each image. You will be asked to send us your translated object based on our camera parameters. You solution will be compared to our ground truth (collected by a 3D scanner). The rotation and translation of our ground truth object are manually labeled to align with the object in the image. </p>\n<p>For the final evaluation, we will take the minimum chamfer distance produced by three kinds of translation. Each translation will be computed to translate your object to the ground truth object. These translations are (1) no translation, (2) translation by Point-to-point ICP from open3D library, (3) translation by Point-to-plane ICP from open3D library. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2805859": "Comparing meshes using Chamfer distance is helpful, but the distance may be significantly higher if the meshes (i.e., the predicted mesh against ground truth mesh) are not aligned due to rotations or translations. From which angle should we present the predicted meshes? From the top, similar to the overhead photos?",
    "2809756": "Once we pass the first phase, we will be providing our evaluation camera parameters for each image. You will be asked to send us your translated object based on our camera parameters. You solution will be compared to our ground truth (collected by a 3D scanner). The rotation and translation of our ground truth object are manually labeled to align with the object in the image. \n\nFor the final evaluation, we will take the minimum chamfer distance produced by three kinds of translation. Each translation will be computed to translate your object to the ground truth object. These translations are (1) no translation, (2) translation by Point-to-point ICP from open3D library, (3) translation by Point-to-plane ICP from open3D library."
  },
  "source": "meta"
}