{
  "id": 174551,
  "title": "TF TPU with JPEGS?",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/174551",
  "author_name": "",
  "post_date": "2020-08-14T04:28:37.156066700Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Did anyone used tensorflow tpu with JPEGS? Please share any public notebook if possible!</p>",
  "messages": [
    {
      "id": "969931",
      "postDate": "08/14/2020 04:28:37",
      "content": "<p>Did anyone used tensorflow tpu with JPEGS? Please share any public notebook if possible!</p>",
      "rawMarkdown": "Did anyone used tensorflow tpu with JPEGS? Please share any public notebook if possible!",
      "votes": null
    },
    {
      "id": "969940",
      "postDate": "08/14/2020 04:44:51",
      "content": "<p>Why would you? Its not as efficient as tfrecords. </p>",
      "rawMarkdown": "Why would you? Its not as efficient as tfrecords.",
      "votes": null
    },
    {
      "id": "969958",
      "postDate": "08/14/2020 04:58:29",
      "content": "<p>I did all of my work with TFRecords but was thinking of using my 1024x1024 jpeg dataset which I created at the starting of the competition, I believe it will give a lot of boost to my ensemble setup.</p>\n<p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>",
      "rawMarkdown": "I did all of my work with TFRecords but was thinking of using my 1024x1024 jpeg dataset which I created at the starting of the competition, I believe it will give a lot of boost to my ensemble setup.\n\nTFRecord for 1024 is not available and I don't know how to create one. :/",
      "votes": null
    },
    {
      "id": "969965",
      "postDate": "08/14/2020 05:11:02",
      "content": "<blockquote>\n  <p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> did create 1024x1024 tfrecord datasets, <a href=\"https://www.kaggle.com/cdeotte/melanoma-1024x1024\" target=\"_blank\">comp only</a>, <a href=\"https://www.kaggle.com/cdeotte/isic2019-1024x1024\" target=\"_blank\">external</a>, and <a href=\"https://www.kaggle.com/cdeotte/malignant-v2-1024x1024\" target=\"_blank\">external v2</a></p>",
      "rawMarkdown": "> TFRecord for 1024 is not available and I don't know how to create one. :/\n\n@cdeotte did create 1024x1024 tfrecord datasets, [comp only](https://www.kaggle.com/cdeotte/melanoma-1024x1024), [external](https://www.kaggle.com/cdeotte/isic2019-1024x1024), and [external v2](https://www.kaggle.com/cdeotte/malignant-v2-1024x1024)",
      "votes": null
    },
    {
      "id": "969969",
      "postDate": "08/14/2020 05:13:10",
      "content": "<p>There is a TPU JPEG example <a href=\"https://www.kaggle.com/truonghoang/fork-of-siim-isic-melanoma-starter\" target=\"_blank\">here</a>. But I have created TFRecords for 1024x1024, links below.</p>",
      "rawMarkdown": "There is a TPU JPEG example [here][1]. But I have created TFRecords for 1024x1024, links below.\n\n[1]: https://www.kaggle.com/truonghoang/fork-of-siim-isic-melanoma-starter",
      "votes": null
    },
    {
      "id": "969971",
      "postDate": "08/14/2020 05:14:54",
      "content": "<blockquote>\n  <p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>\n</blockquote>\n<p>I did create a 1024x1024 TFRecord for 2020, 2019, 2018, 2017, and malignant. <a href=\"https://www.kaggle.com/spoon69\" target=\"_blank\">@spoon69</a> has posted the links above. I also posted an example notebook that uses TPU with JPEG above too.</p>",
      "rawMarkdown": "> TFRecord for 1024 is not available and I don't know how to create one. :/\n\nI did create a 1024x1024 TFRecord for 2020, 2019, 2018, 2017, and malignant. @spoon69 has posted the links above. I also posted an example notebook that uses TPU with JPEG above too.",
      "votes": null
    },
    {
      "id": "970001",
      "postDate": "08/14/2020 05:45:37",
      "content": "<p>Ohh, sorry <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> <a href=\"https://www.kaggle.com/spoon69\" target=\"_blank\">@spoon69</a> I don't know how I missed it, thanks a lot for this!</p>",
      "rawMarkdown": "Ohh, sorry @cdeotte @spoon69 I don't know how I missed it, thanks a lot for this!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 969940,
      "author_name": "teeyee314",
      "author_url": "",
      "post_date": "08/14/2020 04:44:51",
      "content": "<p>Why would you? Its not as efficient as tfrecords. </p>",
      "votes": null,
      "replies": [
        {
          "id": 969958,
          "author_name": "sarques",
          "author_url": "",
          "post_date": "08/14/2020 04:58:29",
          "content": "<p>I did all of my work with TFRecords but was thinking of using my 1024x1024 jpeg dataset which I created at the starting of the competition, I believe it will give a lot of boost to my ensemble setup.</p>\n<p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 969965,
          "author_name": "spoon69",
          "author_url": "",
          "post_date": "08/14/2020 05:11:02",
          "content": "<blockquote>\n  <p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> did create 1024x1024 tfrecord datasets, <a href=\"https://www.kaggle.com/cdeotte/melanoma-1024x1024\" target=\"_blank\">comp only</a>, <a href=\"https://www.kaggle.com/cdeotte/isic2019-1024x1024\" target=\"_blank\">external</a>, and <a href=\"https://www.kaggle.com/cdeotte/malignant-v2-1024x1024\" target=\"_blank\">external v2</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 969971,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "08/14/2020 05:14:54",
          "content": "<blockquote>\n  <p>TFRecord for 1024 is not available and I don't know how to create one. :/</p>\n</blockquote>\n<p>I did create a 1024x1024 TFRecord for 2020, 2019, 2018, 2017, and malignant. <a href=\"https://www.kaggle.com/spoon69\" target=\"_blank\">@spoon69</a> has posted the links above. I also posted an example notebook that uses TPU with JPEG above too.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 970001,
          "author_name": "sarques",
          "author_url": "",
          "post_date": "08/14/2020 05:45:37",
          "content": "<p>Ohh, sorry <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> <a href=\"https://www.kaggle.com/spoon69\" target=\"_blank\">@spoon69</a> I don't know how I missed it, thanks a lot for this!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 969969,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "08/14/2020 05:13:10",
      "content": "<p>There is a TPU JPEG example <a href=\"https://www.kaggle.com/truonghoang/fork-of-siim-isic-melanoma-starter\" target=\"_blank\">here</a>. But I have created TFRecords for 1024x1024, links below.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "969931": "Did anyone used tensorflow tpu with JPEGS? Please share any public notebook if possible!",
    "969940": "Why would you? Its not as efficient as tfrecords.",
    "969958": "I did all of my work with TFRecords but was thinking of using my 1024x1024 jpeg dataset which I created at the starting of the competition, I believe it will give a lot of boost to my ensemble setup.\n\nTFRecord for 1024 is not available and I don't know how to create one. :/",
    "969965": "> TFRecord for 1024 is not available and I don't know how to create one. :/\n\n@cdeotte did create 1024x1024 tfrecord datasets, [comp only](https://www.kaggle.com/cdeotte/melanoma-1024x1024), [external](https://www.kaggle.com/cdeotte/isic2019-1024x1024), and [external v2](https://www.kaggle.com/cdeotte/malignant-v2-1024x1024)",
    "969969": "There is a TPU JPEG example [here][1]. But I have created TFRecords for 1024x1024, links below.\n\n[1]: https://www.kaggle.com/truonghoang/fork-of-siim-isic-melanoma-starter",
    "969971": "> TFRecord for 1024 is not available and I don't know how to create one. :/\n\nI did create a 1024x1024 TFRecord for 2020, 2019, 2018, 2017, and malignant. @spoon69 has posted the links above. I also posted an example notebook that uses TPU with JPEG above too.",
    "970001": "Ohh, sorry @cdeotte @spoon69 I don't know how I missed it, thanks a lot for this!"
  },
  "source": "meta"
}