{
  "id": 207057,
  "title": "Albumentations with TPU",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/207057",
  "author_name": "",
  "post_date": "2020-12-28T01:15:28.729830Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi, does anyone know if augmentations from albumentations can be used with TPUs? I am getting an error that says “Failed to connect to all addresses”.</p>",
  "messages": [
    {
      "id": "1128994",
      "postDate": "12/28/2020 01:15:28",
      "content": "<p>Hi, does anyone know if augmentations from albumentations can be used with TPUs? I am getting an error that says “Failed to connect to all addresses”.</p>",
      "rawMarkdown": "Hi, does anyone know if augmentations from albumentations can be used with TPUs? I am getting an error that says “Failed to connect to all addresses”.",
      "votes": null
    },
    {
      "id": "1129944",
      "postDate": "12/28/2020 17:04:43",
      "content": "<p>I don't think we can use albumentations with TPU.<br>\nOn TPU we are dealing with Tfrecords which are combinations of data stored in Byte format. So , albumentations can't be applied as such there. Though there are many different notebook available which provide multiple augmentations on TPU , you can check them and apply the one you want to experiment.</p>",
      "rawMarkdown": "I don't think we can use albumentations with TPU.\nOn TPU we are dealing with Tfrecords which are combinations of data stored in Byte format. So , albumentations can't be applied as such there. Though there are many different notebook available which provide multiple augmentations on TPU , you can check them and apply the one you want to experiment.",
      "votes": null
    },
    {
      "id": "1130008",
      "postDate": "12/28/2020 17:39:06",
      "content": "<p>Hi,</p>\n<p>In a different competition, there is a notebook which uses TPUs and albumentations together in Pytorch. You can check it from <a href=\"https://www.kaggle.com/joshi98kishan/foldtraining-pytorch-tpu-8-cores\" target=\"_blank\">here</a>. It's pretty straightforward to adapt it to this competition. Changing the model and the dataset class would be sufficient.</p>",
      "rawMarkdown": "Hi,\n\nIn a different competition, there is a notebook which uses TPUs and albumentations together in Pytorch. You can check it from [here](https://www.kaggle.com/joshi98kishan/foldtraining-pytorch-tpu-8-cores). It's pretty straightforward to adapt it to this competition. Changing the model and the dataset class would be sufficient.",
      "votes": null
    },
    {
      "id": "1130065",
      "postDate": "12/28/2020 18:16:01",
      "content": "<p>The question should be \"Albumentations with TFRecords\" and not with TPU.</p>\n<p>You can use Albumentations with TPU , either with Pytorch/XLA or with Keras generator </p>",
      "rawMarkdown": "The question should be \"Albumentations with TFRecords\" and not with TPU.\n\nYou can use Albumentations with TPU , either with Pytorch/XLA or with Keras generator",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1129944,
      "author_name": "prashantarorat",
      "author_url": "",
      "post_date": "12/28/2020 17:04:43",
      "content": "<p>I don't think we can use albumentations with TPU.<br>\nOn TPU we are dealing with Tfrecords which are combinations of data stored in Byte format. So , albumentations can't be applied as such there. Though there are many different notebook available which provide multiple augmentations on TPU , you can check them and apply the one you want to experiment.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1130008,
      "author_name": "snnclsr",
      "author_url": "",
      "post_date": "12/28/2020 17:39:06",
      "content": "<p>Hi,</p>\n<p>In a different competition, there is a notebook which uses TPUs and albumentations together in Pytorch. You can check it from <a href=\"https://www.kaggle.com/joshi98kishan/foldtraining-pytorch-tpu-8-cores\" target=\"_blank\">here</a>. It's pretty straightforward to adapt it to this competition. Changing the model and the dataset class would be sufficient.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1130065,
      "author_name": "serigne",
      "author_url": "",
      "post_date": "12/28/2020 18:16:01",
      "content": "<p>The question should be \"Albumentations with TFRecords\" and not with TPU.</p>\n<p>You can use Albumentations with TPU , either with Pytorch/XLA or with Keras generator </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1128994": "Hi, does anyone know if augmentations from albumentations can be used with TPUs? I am getting an error that says “Failed to connect to all addresses”.",
    "1129944": "I don't think we can use albumentations with TPU.\nOn TPU we are dealing with Tfrecords which are combinations of data stored in Byte format. So , albumentations can't be applied as such there. Though there are many different notebook available which provide multiple augmentations on TPU , you can check them and apply the one you want to experiment.",
    "1130008": "Hi,\n\nIn a different competition, there is a notebook which uses TPUs and albumentations together in Pytorch. You can check it from [here](https://www.kaggle.com/joshi98kishan/foldtraining-pytorch-tpu-8-cores). It's pretty straightforward to adapt it to this competition. Changing the model and the dataset class would be sufficient.",
    "1130065": "The question should be \"Albumentations with TFRecords\" and not with TPU.\n\nYou can use Albumentations with TPU , either with Pytorch/XLA or with Keras generator"
  },
  "source": "meta"
}