{
  "id": 396443,
  "title": "Question on predicted mask shape",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/396443",
  "author_name": "",
  "post_date": "2023-03-21T16:30:02.504901900Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p><a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> should the predicted mask have shape <code>(height,width,65)</code> or <code>(height,width,1)</code> before converting to RLE?</p>\n<p>My question is whether we need to predict a mask for each of the 65 views of the surface volume or 1 single mask for the whole fragment. </p>",
  "messages": [
    {
      "id": "2191004",
      "postDate": "03/21/2023 16:30:02",
      "content": "<p><a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> should the predicted mask have shape <code>(height,width,65)</code> or <code>(height,width,1)</code> before converting to RLE?</p>\n<p>My question is whether we need to predict a mask for each of the 65 views of the surface volume or 1 single mask for the whole fragment. </p>",
      "rawMarkdown": "jpposma should the predicted mask have shape `(height,width,65)` or `(height,width,1)` before converting to RLE?\n\nMy question is whether we need to predict a mask for each of the 65 views of the surface volume or 1 single mask for the whole fragment.",
      "votes": null
    },
    {
      "id": "2191169",
      "postDate": "03/21/2023 19:05:00",
      "content": "<p>from my understanding we are given surface volume data/images (3d) but with 2d masks as ground truth, <br>\nso to answer your question \"we need to predict 1 single mask for the whole fragment\". <br>\nHosts may confirm/correct me if I'm wrong</p>",
      "rawMarkdown": "from my understanding we are given surface volume data/images (3d) but with 2d masks as ground truth, \nso to answer your question \"we need to predict 1 single mask for the whole fragment\". \nHosts may confirm/correct me if I'm wrong",
      "votes": null
    },
    {
      "id": "2191178",
      "postDate": "03/21/2023 19:12:13",
      "content": "<p>This is correct. 🙌🏻</p>",
      "rawMarkdown": "This is correct. 🙌🏻",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2191169,
      "author_name": "imeintanis",
      "author_url": "",
      "post_date": "03/21/2023 19:05:00",
      "content": "<p>from my understanding we are given surface volume data/images (3d) but with 2d masks as ground truth, <br>\nso to answer your question \"we need to predict 1 single mask for the whole fragment\". <br>\nHosts may confirm/correct me if I'm wrong</p>",
      "votes": null,
      "replies": [
        {
          "id": 2191178,
          "author_name": "jpposma",
          "author_url": "",
          "post_date": "03/21/2023 19:12:13",
          "content": "<p>This is correct. 🙌🏻</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2191004": "jpposma should the predicted mask have shape `(height,width,65)` or `(height,width,1)` before converting to RLE?\n\nMy question is whether we need to predict a mask for each of the 65 views of the surface volume or 1 single mask for the whole fragment.",
    "2191169": "from my understanding we are given surface volume data/images (3d) but with 2d masks as ground truth, \nso to answer your question \"we need to predict 1 single mask for the whole fragment\". \nHosts may confirm/correct me if I'm wrong",
    "2191178": "This is correct. 🙌🏻"
  },
  "source": "meta"
}