{
  "id": 408128,
  "title": "Resize strategy",
  "url": "/competitions/image-matching-challenge-2023/discussion/408128",
  "author_name": "",
  "post_date": "2023-05-09T15:47:02.028494Z",
  "votes": 8,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi to all participants! I'm trying to figure out the best way to downscale the images and curious to know what strategy best fits in this challenge. When dealing with keypoints detection/matching and 3d reconstruction there are plenty of methods existing but it's hard/impossible to fit full size images. I'm playing with different resize/downscale methods, some of them work better for one scenes, some of them for others. The methods I currently experimenting with is \"downscale with the same aspect ratio\", \"downscale to square\", \"downscale and pad to square\". From the point of view of different keypoint detection methods, is it bad not to keep the initial aspect ratio? Or is it bad if we have padding? Or maybe all of this stuff doesn't matter and the result will be kind of the same in general? Thanks!</p>",
  "messages": [
    {
      "id": "2251738",
      "postDate": "05/09/2023 15:47:02",
      "content": "<p>Hi to all participants! I'm trying to figure out the best way to downscale the images and curious to know what strategy best fits in this challenge. When dealing with keypoints detection/matching and 3d reconstruction there are plenty of methods existing but it's hard/impossible to fit full size images. I'm playing with different resize/downscale methods, some of them work better for one scenes, some of them for others. The methods I currently experimenting with is \"downscale with the same aspect ratio\", \"downscale to square\", \"downscale and pad to square\". From the point of view of different keypoint detection methods, is it bad not to keep the initial aspect ratio? Or is it bad if we have padding? Or maybe all of this stuff doesn't matter and the result will be kind of the same in general? Thanks!</p>",
      "rawMarkdown": "Hi to all participants! I'm trying to figure out the best way to downscale the images and curious to know what strategy best fits in this challenge. When dealing with keypoints detection/matching and 3d reconstruction there are plenty of methods existing but it's hard/impossible to fit full size images. I'm playing with different resize/downscale methods, some of them work better for one scenes, some of them for others. The methods I currently experimenting with is \"downscale with the same aspect ratio\", \"downscale to square\", \"downscale and pad to square\". From the point of view of different keypoint detection methods, is it bad not to keep the initial aspect ratio? Or is it bad if we have padding? Or maybe all of this stuff doesn't matter and the result will be kind of the same in general? Thanks!",
      "votes": null
    },
    {
      "id": "2251874",
      "postDate": "05/09/2023 17:20:38",
      "content": "<p>I think resizing while preserving the aspect ratio makes more sense. LoFTR also works better that way.</p>",
      "rawMarkdown": "I think resizing while preserving the aspect ratio makes more sense. LoFTR also works better that way.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2251874,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "05/09/2023 17:20:38",
      "content": "<p>I think resizing while preserving the aspect ratio makes more sense. LoFTR also works better that way.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2251738": "Hi to all participants! I'm trying to figure out the best way to downscale the images and curious to know what strategy best fits in this challenge. When dealing with keypoints detection/matching and 3d reconstruction there are plenty of methods existing but it's hard/impossible to fit full size images. I'm playing with different resize/downscale methods, some of them work better for one scenes, some of them for others. The methods I currently experimenting with is \"downscale with the same aspect ratio\", \"downscale to square\", \"downscale and pad to square\". From the point of view of different keypoint detection methods, is it bad not to keep the initial aspect ratio? Or is it bad if we have padding? Or maybe all of this stuff doesn't matter and the result will be kind of the same in general? Thanks!",
    "2251874": "I think resizing while preserving the aspect ratio makes more sense. LoFTR also works better that way."
  },
  "source": "meta"
}