{
  "id": 189693,
  "title": "Role of Instance Normalization / Batch Normalization in Downsample ? ",
  "url": "/competitions/gan-getting-started/discussion/189693",
  "author_name": "Aditya Baurai",
  "post_date": "2020-10-08T10:55:33.787000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello all,</p>\n<p>Quite recently I came across several kernels and some code available on Tensorflow 2.0 guide, focusing on downsample and upsample ops. I have a question : </p>\n<p>Most of these resources use Instance/Batch Normalization only in downsampling(no upsampling). In upsampling, they have used DROPOUT. Why are we doing this?? Can't we use Instance/Batch norm in upsample too? Or Dropout in down? Is it some sort of standard template? </p>",
  "messages": [
    {
      "id": 1042602,
      "postDate": "2020-10-08T10:55:33.787Z",
      "content": "<p>Hello all,</p>\n<p>Quite recently I came across several kernels and some code available on Tensorflow 2.0 guide, focusing on downsample and upsample ops. I have a question : </p>\n<p>Most of these resources use Instance/Batch Normalization only in downsampling(no upsampling). In upsampling, they have used DROPOUT. Why are we doing this?? Can't we use Instance/Batch norm in upsample too? Or Dropout in down? Is it some sort of standard template? </p>",
      "rawMarkdown": "Hello all,\n\nQuite recently I came across several kernels and some code available on Tensorflow 2.0 guide, focusing on downsample and upsample ops. I have a question : \n\nMost of these resources use Instance/Batch Normalization only in downsampling(no upsampling). In upsampling, they have used DROPOUT. Why are we doing this?? Can't we use Instance/Batch norm in upsample too? Or Dropout in down? Is it some sort of standard template? ",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1042602": "Hello all,\n\nQuite recently I came across several kernels and some code available on Tensorflow 2.0 guide, focusing on downsample and upsample ops. I have a question : \n\nMost of these resources use Instance/Batch Normalization only in downsampling(no upsampling). In upsampling, they have used DROPOUT. Why are we doing this?? Can't we use Instance/Batch norm in upsample too? Or Dropout in down? Is it some sort of standard template? "
  }
}