{
  "id": 52206,
  "title": "Multichannel U-net",
  "url": "/competitions/data-science-bowl-2018/discussion/52206",
  "author_name": "Eduardo Rocha de Andrade",
  "post_date": "2018-03-17T13:33:59.537000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi, I'm quite new to Kaggle and DataScience as well and I'm trying to implement a multichannel U-net to detect boundaries, nucleus and background. However, I'm having some problem when modifying the masks to get a multichannel ground-truth.</p>\n\n<p>My prediction must be in the form of [IMG_WIDTH, IMG_HEIGHT, 3], right? So, I have to modify my masks to be in the same form ([IMG_WIDTH, IMG_HEIGHT, 3]) ?</p>\n\n<p>This means that one of the channels (background) must be the negative other the other two combined?</p>\n\n<p>I ask this because I managed to process my masks to get a single channel image, where each pixel can have three values (0, 0.5, 1), which correspond to the classes (background, boundary, inside). Is there any way that I can use this as input?</p>\n\n<p>Thanks in advance</p>",
  "messages": [
    {
      "id": 297589,
      "postDate": "2018-03-17T13:33:59.537Z",
      "content": "<p>Hi, I'm quite new to Kaggle and DataScience as well and I'm trying to implement a multichannel U-net to detect boundaries, nucleus and background. However, I'm having some problem when modifying the masks to get a multichannel ground-truth.</p>\n\n<p>My prediction must be in the form of [IMG_WIDTH, IMG_HEIGHT, 3], right? So, I have to modify my masks to be in the same form ([IMG_WIDTH, IMG_HEIGHT, 3]) ?</p>\n\n<p>This means that one of the channels (background) must be the negative other the other two combined?</p>\n\n<p>I ask this because I managed to process my masks to get a single channel image, where each pixel can have three values (0, 0.5, 1), which correspond to the classes (background, boundary, inside). Is there any way that I can use this as input?</p>\n\n<p>Thanks in advance</p>",
      "rawMarkdown": "Hi, I'm quite new to Kaggle and DataScience as well and I'm trying to implement a multichannel U-net to detect boundaries, nucleus and background. However, I'm having some problem when modifying the masks to get a multichannel ground-truth.\n\nMy prediction must be in the form of [IMG_WIDTH, IMG_HEIGHT, 3], right? So, I have to modify my masks to be in the same form ([IMG_WIDTH, IMG_HEIGHT, 3]) ?\n\nThis means that one of the channels (background) must be the negative other the other two combined?\n\nI ask this because I managed to process my masks to get a single channel image, where each pixel can have three values (0, 0.5, 1), which correspond to the classes (background, boundary, inside). Is there any way that I can use this as input?\n\nThanks in advance"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "297589": "Hi, I'm quite new to Kaggle and DataScience as well and I'm trying to implement a multichannel U-net to detect boundaries, nucleus and background. However, I'm having some problem when modifying the masks to get a multichannel ground-truth.\n\nMy prediction must be in the form of [IMG_WIDTH, IMG_HEIGHT, 3], right? So, I have to modify my masks to be in the same form ([IMG_WIDTH, IMG_HEIGHT, 3]) ?\n\nThis means that one of the channels (background) must be the negative other the other two combined?\n\nI ask this because I managed to process my masks to get a single channel image, where each pixel can have three values (0, 0.5, 1), which correspond to the classes (background, boundary, inside). Is there any way that I can use this as input?\n\nThanks in advance"
  }
}