{
  "id": 161312,
  "title": "Batch norm problems",
  "url": "/competitions/alaska2-image-steganalysis/discussion/161312",
  "author_name": "",
  "post_date": "2020-06-24T14:40:42.688926900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>It seems that the running mean and variance computed during training do not generalize well at all since I get terrible results during evaluation. Anyone else getting the same problem?</p>",
  "messages": [
    {
      "id": "899950",
      "postDate": "06/24/2020 14:40:42",
      "content": "<p>It seems that the running mean and variance computed during training do not generalize well at all since I get terrible results during evaluation. Anyone else getting the same problem?</p>",
      "rawMarkdown": "It seems that the running mean and variance computed during training do not generalize well at all since I get terrible results during evaluation. Anyone else getting the same problem?",
      "votes": null
    },
    {
      "id": "900104",
      "postDate": "06/24/2020 16:14:32",
      "content": "<p>Could it be due to the fact that I am using a cover and a stego of the same image within the same batch during training?</p>",
      "rawMarkdown": "Could it be due to the fact that I am using a cover and a stego of the same image within the same batch during training?",
      "votes": null
    },
    {
      "id": "904974",
      "postDate": "06/28/2020 06:20:18",
      "content": "<p>Is that matter? feed the data with cover, stego in a batch? \nOr just random shuffle?</p>",
      "rawMarkdown": "Is that matter? feed the data with cover, stego in a batch? \nOr just random shuffle?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 900104,
      "author_name": "noachache",
      "author_url": "",
      "post_date": "06/24/2020 16:14:32",
      "content": "<p>Could it be due to the fact that I am using a cover and a stego of the same image within the same batch during training?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 904974,
      "author_name": "cooolz",
      "author_url": "",
      "post_date": "06/28/2020 06:20:18",
      "content": "<p>Is that matter? feed the data with cover, stego in a batch? \nOr just random shuffle?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "899950": "It seems that the running mean and variance computed during training do not generalize well at all since I get terrible results during evaluation. Anyone else getting the same problem?",
    "900104": "Could it be due to the fact that I am using a cover and a stego of the same image within the same batch during training?",
    "904974": "Is that matter? feed the data with cover, stego in a batch? \nOr just random shuffle?"
  },
  "source": "meta"
}