{
  "id": 671790,
  "title": "As a beginner here, I am confused. Help me",
  "url": "/competitions/vesuvius-challenge-surface-detection/discussion/671790",
  "author_name": "",
  "post_date": "2026-02-04T03:57:12.972003600Z",
  "votes": 1,
  "comment_count": 7,
  "views": 0,
  "content": "<p>So after looking the dataset, reading about the competition and few notebooks, I am kinda confused what to do?</p>\n<ul>\n<li><p>This competition talks about 3d volumetric voxels, and predicting writing area in those voxels, but the dataset given in this competition are simple 2D images. </p></li>\n<li><p>What exactly we have to submit? Do we have to submit 3D voxels? or 2D images as given in the input dataset? </p></li>\n<li><p>Do we have to necessarily create 3D volume using images (slices) given in the dataset? And the predict and submit 3D masks? </p></li>\n<li><p>Can we submit simple images here? </p></li>\n<li><p>What can be the starter base-line here (looking at dataset images correctly and also 3d voxels if possible), and predict. </p></li>\n<li><p>After starter base-line, how can we improve our model? Can anyone share? </p></li>\n</ul>",
  "messages": [
    {
      "id": "3401580",
      "postDate": "02/04/2026 03:57:12",
      "content": "<p>So after looking the dataset, reading about the competition and few notebooks, I am kinda confused what to do?</p>\n<ul>\n<li><p>This competition talks about 3d volumetric voxels, and predicting writing area in those voxels, but the dataset given in this competition are simple 2D images. </p></li>\n<li><p>What exactly we have to submit? Do we have to submit 3D voxels? or 2D images as given in the input dataset? </p></li>\n<li><p>Do we have to necessarily create 3D volume using images (slices) given in the dataset? And the predict and submit 3D masks? </p></li>\n<li><p>Can we submit simple images here? </p></li>\n<li><p>What can be the starter base-line here (looking at dataset images correctly and also 3d voxels if possible), and predict. </p></li>\n<li><p>After starter base-line, how can we improve our model? Can anyone share? </p></li>\n</ul>",
      "rawMarkdown": "So after looking the dataset, reading about the competition and few notebooks, I am kinda confused what to do?\n\n- This competition talks about 3d volumetric voxels, and predicting writing area in those voxels, but the dataset given in this competition are simple 2D images. \n\n- What exactly we have to submit? Do we have to submit 3D voxels? or 2D images as given in the input dataset? \n\n- Do we have to necessarily create 3D volume using images (slices) given in the dataset? And the predict and submit 3D masks? \n\n- Can we submit simple images here? \n\n- What can be the starter base-line here (looking at dataset images correctly and also 3d voxels if possible), and predict. \n\n- After starter base-line, how can we improve our model? Can anyone share?",
      "votes": null
    },
    {
      "id": "3401619",
      "postDate": "02/04/2026 06:25:00",
      "content": "<p>Heyy bro, want to connect with you I also have some confusion</p>",
      "rawMarkdown": "Heyy bro, want to connect with you I also have some confusion",
      "votes": null
    },
    {
      "id": "3401655",
      "postDate": "02/04/2026 08:28:03",
      "content": "<p>• The game data comes in two formats: 256×256×256 and 320×320×320, not 2D.</p>\n<p>• Submit 3D files, like LABEL does.</p>\n<p>• As long as it conforms to the LABEL format, it can be processed in any way.</p>\n<p>• BASELINE and optimization strategies can be found in the discussion forum, helping you quickly get started and reach the Bronze tier.</p>",
      "rawMarkdown": "• The game data comes in two formats: 256×256×256 and 320×320×320, not 2D.\n\n• Submit 3D files, like LABEL does.\n\n• As long as it conforms to the LABEL format, it can be processed in any way.\n\n• BASELINE and optimization strategies can be found in the discussion forum, helping you quickly get started and reach the Bronze tier.",
      "votes": null
    },
    {
      "id": "3401703",
      "postDate": "02/04/2026 10:57:26",
      "content": "<p>input data is 3D \nuse tifffile.imread to get the npy \nlabels are of 3 classes 0,1,2 (2 being ignore)</p>",
      "rawMarkdown": "input data is 3D \nuse tifffile.imread to get the npy \nlabels are of 3 classes 0,1,2 (2 being ignore)",
      "votes": null
    },
    {
      "id": "3401720",
      "postDate": "02/04/2026 11:37:08",
      "content": "<p>you said input data is 3d but Inside train_images, test_images and labels directory, there are only images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model?</p>",
      "rawMarkdown": "you said input data is 3d but Inside train_images, test_images and labels directory, there are only images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model?",
      "votes": null
    },
    {
      "id": "3401721",
      "postDate": "02/04/2026 11:38:33",
      "content": "<p>Inside train_images, test_images and labels directory, there are only 2d images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model? label directory also has 2d images rather than 3d models.</p>",
      "rawMarkdown": "Inside train_images, test_images and labels directory, there are only 2d images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model? label directory also has 2d images rather than 3d models.",
      "votes": null
    },
    {
      "id": "3401724",
      "postDate": "02/04/2026 11:58:16",
      "content": "<p>bro do img = tifffile.imread(path)\nimg.shape</p>",
      "rawMarkdown": "bro do img = tifffile.imread(path)\nimg.shape",
      "votes": null
    },
    {
      "id": "3401727",
      "postDate": "02/04/2026 12:08:23",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F15255201%2F315425265a519488319f1fb96157671b%2FScreenshot%202026-02-04%20200737.png?generation=1770206888398374&amp;alt=media\" alt=\"\"></p>\n<p>320x320x320</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F15255201%2F315425265a519488319f1fb96157671b%2FScreenshot%202026-02-04%20200737.png?generation=1770206888398374&alt=media)\n\n320x320x320",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3401619,
      "author_name": "aknexa",
      "author_url": "",
      "post_date": "02/04/2026 06:25:00",
      "content": "<p>Heyy bro, want to connect with you I also have some confusion</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3401655,
      "author_name": "ggayoayogg",
      "author_url": "",
      "post_date": "02/04/2026 08:28:03",
      "content": "<p>• The game data comes in two formats: 256×256×256 and 320×320×320, not 2D.</p>\n<p>• Submit 3D files, like LABEL does.</p>\n<p>• As long as it conforms to the LABEL format, it can be processed in any way.</p>\n<p>• BASELINE and optimization strategies can be found in the discussion forum, helping you quickly get started and reach the Bronze tier.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3401721,
          "author_name": "luckyslodhi",
          "author_url": "",
          "post_date": "02/04/2026 11:38:33",
          "content": "<p>Inside train_images, test_images and labels directory, there are only 2d images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model? label directory also has 2d images rather than 3d models.</p>",
          "votes": null,
          "replies": [
            {
              "id": 3401727,
              "author_name": "ggayoayogg",
              "author_url": "",
              "post_date": "02/04/2026 12:08:23",
              "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F15255201%2F315425265a519488319f1fb96157671b%2FScreenshot%202026-02-04%20200737.png?generation=1770206888398374&amp;alt=media\" alt=\"\"></p>\n<p>320x320x320</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 3401703,
      "author_name": "arjunashokbhandary",
      "author_url": "",
      "post_date": "02/04/2026 10:57:26",
      "content": "<p>input data is 3D \nuse tifffile.imread to get the npy \nlabels are of 3 classes 0,1,2 (2 being ignore)</p>",
      "votes": null,
      "replies": [
        {
          "id": 3401720,
          "author_name": "luckyslodhi",
          "author_url": "",
          "post_date": "02/04/2026 11:37:08",
          "content": "<p>you said input data is 3d but Inside train_images, test_images and labels directory, there are only images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model?</p>",
          "votes": null,
          "replies": [
            {
              "id": 3401724,
              "author_name": "arjunashokbhandary",
              "author_url": "",
              "post_date": "02/04/2026 11:58:16",
              "content": "<p>bro do img = tifffile.imread(path)\nimg.shape</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3401580": "So after looking the dataset, reading about the competition and few notebooks, I am kinda confused what to do?\n\n- This competition talks about 3d volumetric voxels, and predicting writing area in those voxels, but the dataset given in this competition are simple 2D images. \n\n- What exactly we have to submit? Do we have to submit 3D voxels? or 2D images as given in the input dataset? \n\n- Do we have to necessarily create 3D volume using images (slices) given in the dataset? And the predict and submit 3D masks? \n\n- Can we submit simple images here? \n\n- What can be the starter base-line here (looking at dataset images correctly and also 3d voxels if possible), and predict. \n\n- After starter base-line, how can we improve our model? Can anyone share?",
    "3401619": "Heyy bro, want to connect with you I also have some confusion",
    "3401655": "• The game data comes in two formats: 256×256×256 and 320×320×320, not 2D.\n\n• Submit 3D files, like LABEL does.\n\n• As long as it conforms to the LABEL format, it can be processed in any way.\n\n• BASELINE and optimization strategies can be found in the discussion forum, helping you quickly get started and reach the Bronze tier.",
    "3401703": "input data is 3D \nuse tifffile.imread to get the npy \nlabels are of 3 classes 0,1,2 (2 being ignore)",
    "3401720": "you said input data is 3d but Inside train_images, test_images and labels directory, there are only images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model?",
    "3401721": "Inside train_images, test_images and labels directory, there are only 2d images with tiff extension. how simple images can be 3d? do we have to stack the images to create a some kind of 3d model? label directory also has 2d images rather than 3d models.",
    "3401724": "bro do img = tifffile.imread(path)\nimg.shape",
    "3401727": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F15255201%2F315425265a519488319f1fb96157671b%2FScreenshot%202026-02-04%20200737.png?generation=1770206888398374&alt=media)\n\n320x320x320"
  },
  "source": "meta"
}