{
  "id": 216958,
  "title": "The best way to load uint8 and uint16",
  "url": "/competitions/hpa-single-cell-image-classification/discussion/216958",
  "author_name": "Vladislav Ostankovich",
  "post_date": "2021-02-04T16:16:53.097000",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi there!</p>\n<p>What is the best way to load the images from train/test sets? Issue is that train set contains uint8 images and test set contains uint16 images (not sure if hidden test set is also uint16). The question is how to load uint8 and uint16 images so that the difference is negligible? I mean, for example, if uint16 image is loaded and then normalized to max255, it looks brighter/sharper that uint8. And what libraries would you suggest for loading uint16 images? Thanks!</p>",
  "messages": [
    {
      "id": 1186180,
      "postDate": "2021-02-04T16:16:53.097Z",
      "content": "<p>Hi there!</p>\n<p>What is the best way to load the images from train/test sets? Issue is that train set contains uint8 images and test set contains uint16 images (not sure if hidden test set is also uint16). The question is how to load uint8 and uint16 images so that the difference is negligible? I mean, for example, if uint16 image is loaded and then normalized to max255, it looks brighter/sharper that uint8. And what libraries would you suggest for loading uint16 images? Thanks!</p>",
      "rawMarkdown": "Hi there!\n\nWhat is the best way to load the images from train/test sets? Issue is that train set contains uint8 images and test set contains uint16 images (not sure if hidden test set is also uint16). The question is how to load uint8 and uint16 images so that the difference is negligible? I mean, for example, if uint16 image is loaded and then normalized to max255, it looks brighter/sharper that uint8. And what libraries would you suggest for loading uint16 images? Thanks!",
      "votes": 4
    },
    {
      "id": 1189794,
      "postDate": "2021-02-07T08:26:17.687Z",
      "content": "<p>I would think that you would just convert them to a common datatype and normalize them so they are equivalent for both cases. I dont know why they would end up brighter. Maybe a tiny bit more precision in the range of values, but I think that would likely not be too noticeable. At least not visually. </p>",
      "rawMarkdown": "I would think that you would just convert them to a common datatype and normalize them so they are equivalent for both cases. I dont know why they would end up brighter. Maybe a tiny bit more precision in the range of values, but I think that would likely not be too noticeable. At least not visually. "
    }
  ],
  "comments": [
    {
      "id": 1189794,
      "author_name": "ryches",
      "author_url": "",
      "post_date": "2021-02-07T08:26:17.687000",
      "content": "<p>I would think that you would just convert them to a common datatype and normalize them so they are equivalent for both cases. I dont know why they would end up brighter. Maybe a tiny bit more precision in the range of values, but I think that would likely not be too noticeable. At least not visually. </p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1186180": "Hi there!\n\nWhat is the best way to load the images from train/test sets? Issue is that train set contains uint8 images and test set contains uint16 images (not sure if hidden test set is also uint16). The question is how to load uint8 and uint16 images so that the difference is negligible? I mean, for example, if uint16 image is loaded and then normalized to max255, it looks brighter/sharper that uint8. And what libraries would you suggest for loading uint16 images? Thanks!",
    "1189794": "I would think that you would just convert them to a common datatype and normalize them so they are equivalent for both cases. I dont know why they would end up brighter. Maybe a tiny bit more precision in the range of values, but I think that would likely not be too noticeable. At least not visually. "
  }
}