{
  "id": 113036,
  "title": "Reduce Data Size",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/113036",
  "author_name": "Shervin",
  "post_date": "2019-10-16T16:20:15.964000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hey Kagglers,</p>\n\n<p>Does anyone know how to reduce the data size?</p>",
  "messages": [
    {
      "id": 650800,
      "postDate": "2019-10-16T17:36:41.900Z",
      "content": "<p>Download .png\n<a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419</a> </p>",
      "rawMarkdown": "Download .png\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419 ",
      "votes": 1,
      "replies": [
        {
          "id": 651201,
          "postDate": "2019-10-17T07:06:36.847Z",
          "content": "<p>Thanks 👌 </p>",
          "rawMarkdown": "Thanks 👌 "
        }
      ]
    },
    {
      "id": 650726,
      "postDate": "2019-10-16T16:20:15.963Z",
      "content": "<p>Hey Kagglers,</p>\n\n<p>Does anyone know how to reduce the data size?</p>",
      "rawMarkdown": "Hey Kagglers,\n\nDoes anyone know how to reduce the data size?",
      "votes": 1
    },
    {
      "id": 651208,
      "postDate": "2019-10-17T07:14:51.073Z",
      "content": "<p>My current solution to reduce the data on my local PC (HDD is way too slow, SSD not big enough), is to use minimal preprocessing on the DICOM data (HU-scaling and Resampling to 1 x 1mm pixels) and save the results as 16bit ints with joblib.dump(compress=3). This reduced the total size to around 40GB from I think 400GB.</p>\n\n<p>PNGs are a lot smaller though, but as Jeremy Howard pointed out in his <a href=\"https://www.kaggle.com/jhoward/don-t-see-like-a-radiologist-fastai\">Don't see like a radiologist kernel </a>, they also compress a lot of information. Maybe windowing is not the way to go in the end, who knows.</p>",
      "rawMarkdown": "My current solution to reduce the data on my local PC (HDD is way too slow, SSD not big enough), is to use minimal preprocessing on the DICOM data (HU-scaling and Resampling to 1 x 1mm pixels) and save the results as 16bit ints with joblib.dump(compress=3). This reduced the total size to around 40GB from I think 400GB.\n\nPNGs are a lot smaller though, but as Jeremy Howard pointed out in his [Don't see like a radiologist kernel ](https://www.kaggle.com/jhoward/don-t-see-like-a-radiologist-fastai), they also compress a lot of information. Maybe windowing is not the way to go in the end, who knows.",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 650800,
      "author_name": "Eek The Cat",
      "author_url": "",
      "post_date": "2019-10-16T17:36:41.900000",
      "content": "<p>Download .png\n<a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419</a> </p>",
      "votes": 1,
      "replies": [
        {
          "id": 651201,
          "author_name": "Shervin",
          "author_url": "",
          "post_date": "2019-10-17T07:06:36.847000",
          "content": "<p>Thanks 👌 </p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 651208,
      "author_name": "srs",
      "author_url": "",
      "post_date": "2019-10-17T07:14:51.073000",
      "content": "<p>My current solution to reduce the data on my local PC (HDD is way too slow, SSD not big enough), is to use minimal preprocessing on the DICOM data (HU-scaling and Resampling to 1 x 1mm pixels) and save the results as 16bit ints with joblib.dump(compress=3). This reduced the total size to around 40GB from I think 400GB.</p>\n\n<p>PNGs are a lot smaller though, but as Jeremy Howard pointed out in his <a href=\"https://www.kaggle.com/jhoward/don-t-see-like-a-radiologist-fastai\">Don't see like a radiologist kernel </a>, they also compress a lot of information. Maybe windowing is not the way to go in the end, who knows.</p>",
      "votes": 2,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "650800": "Download .png\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/110840#latest-650419 ",
    "650726": "Hey Kagglers,\n\nDoes anyone know how to reduce the data size?",
    "651208": "My current solution to reduce the data on my local PC (HDD is way too slow, SSD not big enough), is to use minimal preprocessing on the DICOM data (HU-scaling and Resampling to 1 x 1mm pixels) and save the results as 16bit ints with joblib.dump(compress=3). This reduced the total size to around 40GB from I think 400GB.\n\nPNGs are a lot smaller though, but as Jeremy Howard pointed out in his [Don't see like a radiologist kernel ](https://www.kaggle.com/jhoward/don-t-see-like-a-radiologist-fastai), they also compress a lot of information. Maybe windowing is not the way to go in the end, who knows."
  }
}