{
  "id": 487132,
  "title": "image extensions in train csv file doesn't match with image in folders",
  "url": "/competitions/image-matching-challenge-2024/discussion/487132",
  "author_name": "",
  "post_date": "2024-03-27T18:52:35.103506100Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>While the real image file extension is png, it seems jpeg in the train csv file.<br>\nfor instance     DSC_8243.JPG in train csv file but the extension is png in image folder</p>",
  "messages": [
    {
      "id": "2719488",
      "postDate": "03/27/2024 18:52:35",
      "content": "<p>While the real image file extension is png, it seems jpeg in the train csv file.<br>\nfor instance     DSC_8243.JPG in train csv file but the extension is png in image folder</p>",
      "rawMarkdown": "While the real image file extension is png, it seems jpeg in the train csv file.\nfor instance \tDSC_8243.JPG in train csv file but the extension is png in image folder",
      "votes": null
    },
    {
      "id": "2719599",
      "postDate": "03/27/2024 20:18:57",
      "content": "<p>That's my mistake, sorry. I'll try to have a corrected version of train.csv posted by the end of the week.</p>",
      "rawMarkdown": "That's my mistake, sorry. I'll try to have a corrected version of train.csv posted by the end of the week.",
      "votes": null
    },
    {
      "id": "2719613",
      "postDate": "03/27/2024 20:23:38",
      "content": "<p>thank you Sohier. </p>",
      "rawMarkdown": "thank you Sohier.",
      "votes": null
    },
    {
      "id": "2745068",
      "postDate": "04/10/2024 10:50:56",
      "content": "<p>The images in the sfm images.txt files also don't match. At least in the church scene, the image files are .png, but in the sfm/images.txt they are .jpg. I haven't checked other scenes yet.</p>\n<p>I figured this out because I was trying to use the evaluation metric to check my predictions against the training data reference reconstructions. I slightly altered the code from <a href=\"https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example\" target=\"_blank\">this notebook</a> to take two pycolmap Reconstruction objects directly instead of working off of CSV. At first it was returning 0, so I found that it was not pairing any of the images based on their names.</p>\n<p>This makes me wonder, is something similar going on with the leaderboard? I finally made a submission today and my score is zero. While I wasn't expecting to do great (my code is basically just the example notebook), I was expecting to do better than zero.</p>",
      "rawMarkdown": "The images in the sfm images.txt files also don't match. At least in the church scene, the image files are .png, but in the sfm/images.txt they are .jpg. I haven't checked other scenes yet.\n\nI figured this out because I was trying to use the evaluation metric to check my predictions against the training data reference reconstructions. I slightly altered the code from [this notebook](https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example) to take two pycolmap Reconstruction objects directly instead of working off of CSV. At first it was returning 0, so I found that it was not pairing any of the images based on their names.\n\nThis makes me wonder, is something similar going on with the leaderboard? I finally made a submission today and my score is zero. While I wasn't expecting to do great (my code is basically just the example notebook), I was expecting to do better than zero.",
      "votes": null
    },
    {
      "id": "2776476",
      "postDate": "04/26/2024 07:35:36",
      "content": "<p>There are a lot of inconsistencies in the <code>sfm</code> directories as well. The images in there usually have .jpg extensions instead of .png like in the images directory. Furthermore there are a lot of mistmatches in case. Those two problems are easy enough to fix, but there are some more tricky situations like in the dioscuri sfm, there is an image named <code>templia__75</code> (two underscores) while the image in the <code>images</code> dir is <code>templia_75.png</code> (one underscore).</p>",
      "rawMarkdown": "There are a lot of inconsistencies in the `sfm` directories as well. The images in there usually have .jpg extensions instead of .png like in the images directory. Furthermore there are a lot of mistmatches in case. Those two problems are easy enough to fix, but there are some more tricky situations like in the dioscuri sfm, there is an image named `templia__75` (two underscores) while the image in the `images` dir is `templia_75.png` (one underscore).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2719599,
      "author_name": "sohier",
      "author_url": "",
      "post_date": "03/27/2024 20:18:57",
      "content": "<p>That's my mistake, sorry. I'll try to have a corrected version of train.csv posted by the end of the week.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2719613,
          "author_name": "umar47",
          "author_url": "",
          "post_date": "03/27/2024 20:23:38",
          "content": "<p>thank you Sohier. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2745068,
      "author_name": "zacstewart",
      "author_url": "",
      "post_date": "04/10/2024 10:50:56",
      "content": "<p>The images in the sfm images.txt files also don't match. At least in the church scene, the image files are .png, but in the sfm/images.txt they are .jpg. I haven't checked other scenes yet.</p>\n<p>I figured this out because I was trying to use the evaluation metric to check my predictions against the training data reference reconstructions. I slightly altered the code from <a href=\"https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example\" target=\"_blank\">this notebook</a> to take two pycolmap Reconstruction objects directly instead of working off of CSV. At first it was returning 0, so I found that it was not pairing any of the images based on their names.</p>\n<p>This makes me wonder, is something similar going on with the leaderboard? I finally made a submission today and my score is zero. While I wasn't expecting to do great (my code is basically just the example notebook), I was expecting to do better than zero.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2776476,
      "author_name": "zacstewart",
      "author_url": "",
      "post_date": "04/26/2024 07:35:36",
      "content": "<p>There are a lot of inconsistencies in the <code>sfm</code> directories as well. The images in there usually have .jpg extensions instead of .png like in the images directory. Furthermore there are a lot of mistmatches in case. Those two problems are easy enough to fix, but there are some more tricky situations like in the dioscuri sfm, there is an image named <code>templia__75</code> (two underscores) while the image in the <code>images</code> dir is <code>templia_75.png</code> (one underscore).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2719488": "While the real image file extension is png, it seems jpeg in the train csv file.\nfor instance \tDSC_8243.JPG in train csv file but the extension is png in image folder",
    "2719599": "That's my mistake, sorry. I'll try to have a corrected version of train.csv posted by the end of the week.",
    "2719613": "thank you Sohier.",
    "2745068": "The images in the sfm images.txt files also don't match. At least in the church scene, the image files are .png, but in the sfm/images.txt they are .jpg. I haven't checked other scenes yet.\n\nI figured this out because I was trying to use the evaluation metric to check my predictions against the training data reference reconstructions. I slightly altered the code from [this notebook](https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example) to take two pycolmap Reconstruction objects directly instead of working off of CSV. At first it was returning 0, so I found that it was not pairing any of the images based on their names.\n\nThis makes me wonder, is something similar going on with the leaderboard? I finally made a submission today and my score is zero. While I wasn't expecting to do great (my code is basically just the example notebook), I was expecting to do better than zero.",
    "2776476": "There are a lot of inconsistencies in the `sfm` directories as well. The images in there usually have .jpg extensions instead of .png like in the images directory. Furthermore there are a lot of mistmatches in case. Those two problems are easy enough to fix, but there are some more tricky situations like in the dioscuri sfm, there is an image named `templia__75` (two underscores) while the image in the `images` dir is `templia_75.png` (one underscore)."
  },
  "source": "meta"
}