{
  "id": 132724,
  "title": "Which is better, 3 channel image or 1 channel image?",
  "url": "/competitions/bengaliai-cv19/discussion/132724",
  "author_name": "Yovin Yahathugoda",
  "post_date": "2020-02-27T13:36:20.951000",
  "votes": 0,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I am currently using only single channel images to train, but could someone tell me if using 3 channel images gives a significant boost or is the improvement hardly noticeable?</p>",
  "messages": [
    {
      "id": 758216,
      "postDate": "2020-02-27T14:52:48.453Z",
      "content": "<p>It is most efficient to store, preprocess, and input the images into your model as 1 channel. Once they are \"inside\" your conv net, you can use the first layer to convert the image to 3 channel (either by applying convolutions or simple channel duplication). The advantage of converting to 3 channel is that your model can then proceed to use pretrained ImageNet CNNs inside.</p>",
      "rawMarkdown": "It is most efficient to store, preprocess, and input the images into your model as 1 channel. Once they are \"inside\" your conv net, you can use the first layer to convert the image to 3 channel (either by applying convolutions or simple channel duplication). The advantage of converting to 3 channel is that your model can then proceed to use pretrained ImageNet CNNs inside.",
      "votes": 4
    },
    {
      "id": 758249,
      "postDate": "2020-02-27T15:27:55.847Z",
      "content": "<p>Taken the monstrous fitting capacity of the models, using 3-channel or 1-channels should have minimal, minimal, really minimal effects</p>",
      "rawMarkdown": "Taken the monstrous fitting capacity of the models, using 3-channel or 1-channels should have minimal, minimal, really minimal effects",
      "votes": 1
    },
    {
      "id": 758855,
      "postDate": "2020-02-28T08:58:11.077Z",
      "content": "<p>3 channels have more information, more possibility to extract correct information.</p>",
      "rawMarkdown": "3 channels have more information, more possibility to extract correct information."
    },
    {
      "id": 758767,
      "postDate": "2020-02-28T05:50:27.257Z",
      "content": "<p>Thank you <a href=\"/cdeotte\">@cdeotte</a> and <a href=\"/roguekk007\">@roguekk007</a>  for your replies</p>",
      "rawMarkdown": "Thank you @cdeotte and @roguekk007  for your replies"
    },
    {
      "id": 758148,
      "postDate": "2020-02-27T13:36:20.950Z",
      "content": "<p>I am currently using only single channel images to train, but could someone tell me if using 3 channel images gives a significant boost or is the improvement hardly noticeable?</p>",
      "rawMarkdown": "I am currently using only single channel images to train, but could someone tell me if using 3 channel images gives a significant boost or is the improvement hardly noticeable?"
    },
    {
      "id": 758853,
      "postDate": "2020-02-28T08:57:50.517Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 758216,
      "author_name": "Chris Deotte",
      "author_url": "",
      "post_date": "2020-02-27T14:52:48.453000",
      "content": "<p>It is most efficient to store, preprocess, and input the images into your model as 1 channel. Once they are \"inside\" your conv net, you can use the first layer to convert the image to 3 channel (either by applying convolutions or simple channel duplication). The advantage of converting to 3 channel is that your model can then proceed to use pretrained ImageNet CNNs inside.</p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 758249,
      "author_name": "Nicholas Lyu",
      "author_url": "",
      "post_date": "2020-02-27T15:27:55.847000",
      "content": "<p>Taken the monstrous fitting capacity of the models, using 3-channel or 1-channels should have minimal, minimal, really minimal effects</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 758855,
      "author_name": "Saravanan Chandran",
      "author_url": "",
      "post_date": "2020-02-28T08:58:11.077000",
      "content": "<p>3 channels have more information, more possibility to extract correct information.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 758767,
      "author_name": "Yovin Yahathugoda",
      "author_url": "",
      "post_date": "2020-02-28T05:50:27.257000",
      "content": "<p>Thank you <a href=\"/cdeotte\">@cdeotte</a> and <a href=\"/roguekk007\">@roguekk007</a>  for your replies</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 758853,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-02-28T08:57:50.517000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "758216": "It is most efficient to store, preprocess, and input the images into your model as 1 channel. Once they are \"inside\" your conv net, you can use the first layer to convert the image to 3 channel (either by applying convolutions or simple channel duplication). The advantage of converting to 3 channel is that your model can then proceed to use pretrained ImageNet CNNs inside.",
    "758249": "Taken the monstrous fitting capacity of the models, using 3-channel or 1-channels should have minimal, minimal, really minimal effects",
    "758855": "3 channels have more information, more possibility to extract correct information.",
    "758767": "Thank you @cdeotte and @roguekk007  for your replies",
    "758148": "I am currently using only single channel images to train, but could someone tell me if using 3 channel images gives a significant boost or is the improvement hardly noticeable?",
    "758853": ""
  }
}