{
  "id": 250687,
  "title": "Why albumentations over torchvision.transform for augmentation?",
  "url": "/competitions/seti-breakthrough-listen/discussion/250687",
  "author_name": "Yash Raizada",
  "post_date": "2021-07-03T20:48:59.012000",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I've been reading many extremely helpful notebooks here and observed that almost all these notebooks used albumentations library for all the augmentations. However, I could not understand why it was preferred over the torchvision.transform library. Here's what I found out:</p>\n<p>As per the documentation, in addition to providing a larger number of augmentation methods, the main advantage of albumentations is its super fast computation. As per the comparison in <a href=\"https://github.com/albumentations-team/albumentations#benchmarking-results\" target=\"_blank\">this</a> table, the number of images processed per second is significantly higher in case albumentations than any other library.</p>\n<p>Just thought of sharing to help all the newbies here. Please add if I missed any other benefit. Thanks!</p>",
  "messages": [
    {
      "id": 1375097,
      "postDate": "2021-07-03T20:48:59.013Z",
      "content": "<p>I've been reading many extremely helpful notebooks here and observed that almost all these notebooks used albumentations library for all the augmentations. However, I could not understand why it was preferred over the torchvision.transform library. Here's what I found out:</p>\n<p>As per the documentation, in addition to providing a larger number of augmentation methods, the main advantage of albumentations is its super fast computation. As per the comparison in <a href=\"https://github.com/albumentations-team/albumentations#benchmarking-results\" target=\"_blank\">this</a> table, the number of images processed per second is significantly higher in case albumentations than any other library.</p>\n<p>Just thought of sharing to help all the newbies here. Please add if I missed any other benefit. Thanks!</p>",
      "rawMarkdown": "I've been reading many extremely helpful notebooks here and observed that almost all these notebooks used albumentations library for all the augmentations. However, I could not understand why it was preferred over the torchvision.transform library. Here's what I found out:\n\nAs per the documentation, in addition to providing a larger number of augmentation methods, the main advantage of albumentations is its super fast computation. As per the comparison in [this](https://github.com/albumentations-team/albumentations#benchmarking-results) table, the number of images processed per second is significantly higher in case albumentations than any other library.\n\nJust thought of sharing to help all the newbies here. Please add if I missed any other benefit. Thanks!",
      "votes": 6
    },
    {
      "id": 1378519,
      "postDate": "2021-07-06T15:32:45.193Z",
      "content": "<blockquote>\n  <p>the main advantage of albumentations is its super fast computation</p>\n</blockquote>\n<p>Actually, using GPU is much faster.  Albumentation is the fastest on CPU as shown in the benchmark, but GPU is faster.  Albumentation advantage is the wide range of transforms in a unified API.</p>",
      "rawMarkdown": "> the main advantage of albumentations is its super fast computation\n\nActually, using GPU is much faster.  Albumentation is the fastest on CPU as shown in the benchmark, but GPU is faster.  Albumentation advantage is the wide range of transforms in a unified API.",
      "votes": 3
    },
    {
      "id": 1378840,
      "postDate": "2021-07-06T21:14:05.397Z",
      "content": "<p>If your gpu(s) are already at 100% and your CPU cores have cycles to spare, might as well offload to cpu</p>",
      "rawMarkdown": "If your gpu(s) are already at 100% and your CPU cores have cycles to spare, might as well offload to cpu",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 1378519,
      "author_name": "CPMP",
      "author_url": "",
      "post_date": "2021-07-06T15:32:45.193000",
      "content": "<blockquote>\n  <p>the main advantage of albumentations is its super fast computation</p>\n</blockquote>\n<p>Actually, using GPU is much faster.  Albumentation is the fastest on CPU as shown in the benchmark, but GPU is faster.  Albumentation advantage is the wide range of transforms in a unified API.</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 1378840,
      "author_name": "عثمان",
      "author_url": "",
      "post_date": "2021-07-06T21:14:05.397000",
      "content": "<p>If your gpu(s) are already at 100% and your CPU cores have cycles to spare, might as well offload to cpu</p>",
      "votes": 2,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1375097": "I've been reading many extremely helpful notebooks here and observed that almost all these notebooks used albumentations library for all the augmentations. However, I could not understand why it was preferred over the torchvision.transform library. Here's what I found out:\n\nAs per the documentation, in addition to providing a larger number of augmentation methods, the main advantage of albumentations is its super fast computation. As per the comparison in [this](https://github.com/albumentations-team/albumentations#benchmarking-results) table, the number of images processed per second is significantly higher in case albumentations than any other library.\n\nJust thought of sharing to help all the newbies here. Please add if I missed any other benefit. Thanks!",
    "1378519": "> the main advantage of albumentations is its super fast computation\n\nActually, using GPU is much faster.  Albumentation is the fastest on CPU as shown in the benchmark, but GPU is faster.  Albumentation advantage is the wide range of transforms in a unified API.",
    "1378840": "If your gpu(s) are already at 100% and your CPU cores have cycles to spare, might as well offload to cpu"
  }
}