{
  "id": 104855,
  "title": "Imagenet normalization",
  "url": "/competitions/aptos2019-blindness-detection/discussion/104855",
  "author_name": "Winter",
  "post_date": "2019-08-19T16:12:03.336000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>If we are using pretrained ImageNet models, we should normalize with their mean and std because the pretrained model has learn specifics to that normalization. However after normalization that way the images seem heavily skewed towards one colour channel, is that normal, how can our model train properly like that?</p>",
  "messages": [
    {
      "id": 602899,
      "postDate": "2019-08-19T16:12:03.337Z",
      "content": "<p>If we are using pretrained ImageNet models, we should normalize with their mean and std because the pretrained model has learn specifics to that normalization. However after normalization that way the images seem heavily skewed towards one colour channel, is that normal, how can our model train properly like that?</p>",
      "rawMarkdown": "If we are using pretrained ImageNet models, we should normalize with their mean and std because the pretrained model has learn specifics to that normalization. However after normalization that way the images seem heavily skewed towards one colour channel, is that normal, how can our model train properly like that?",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "602899": "If we are using pretrained ImageNet models, we should normalize with their mean and std because the pretrained model has learn specifics to that normalization. However after normalization that way the images seem heavily skewed towards one colour channel, is that normal, how can our model train properly like that?"
  }
}