{
  "id": 157872,
  "title": "Does Albumentation Slows Training",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/157872",
  "author_name": "Jaideep",
  "post_date": "2020-06-12T10:33:16.795000",
  "votes": 0,
  "comment_count": 1,
  "views": 0,
  "content": "<p>How much is the Epoch Time  with 5 to 7 augmentations used  in Pipeline from package. \nI use image contrast family ones,shift scale, training times increases 4 times with V100 gpu</p>",
  "messages": [
    {
      "id": 883254,
      "postDate": "2020-06-12T13:41:50.233Z",
      "content": "<p>It obviously increases time since it takes longer to create an image... Ideally you want enough CPU power to load images in parallel so that it doesn't become a bottleneck. If your CPU is faster at loading images than your GPU is at iterating then it shouldn't be a problem. You can usually assess this with nvidia-smi during training. If your GPU is often at 0% it means it's waiting for a batch.</p>",
      "rawMarkdown": "It obviously increases time since it takes longer to create an image... Ideally you want enough CPU power to load images in parallel so that it doesn't become a bottleneck. If your CPU is faster at loading images than your GPU is at iterating then it shouldn't be a problem. You can usually assess this with nvidia-smi during training. If your GPU is often at 0% it means it's waiting for a batch."
    },
    {
      "id": 883056,
      "postDate": "2020-06-12T10:33:16.797Z",
      "content": "<p>How much is the Epoch Time  with 5 to 7 augmentations used  in Pipeline from package. \nI use image contrast family ones,shift scale, training times increases 4 times with V100 gpu</p>",
      "rawMarkdown": "How much is the Epoch Time  with 5 to 7 augmentations used  in Pipeline from package. \nI use image contrast family ones,shift scale, training times increases 4 times with V100 gpu\n"
    }
  ],
  "comments": [
    {
      "id": 883254,
      "author_name": "Arnaud Roussel",
      "author_url": "",
      "post_date": "2020-06-12T13:41:50.233000",
      "content": "<p>It obviously increases time since it takes longer to create an image... Ideally you want enough CPU power to load images in parallel so that it doesn't become a bottleneck. If your CPU is faster at loading images than your GPU is at iterating then it shouldn't be a problem. You can usually assess this with nvidia-smi during training. If your GPU is often at 0% it means it's waiting for a batch.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "883254": "It obviously increases time since it takes longer to create an image... Ideally you want enough CPU power to load images in parallel so that it doesn't become a bottleneck. If your CPU is faster at loading images than your GPU is at iterating then it shouldn't be a problem. You can usually assess this with nvidia-smi during training. If your GPU is often at 0% it means it's waiting for a batch.",
    "883056": "How much is the Epoch Time  with 5 to 7 augmentations used  in Pipeline from package. \nI use image contrast family ones,shift scale, training times increases 4 times with V100 gpu\n"
  }
}