{
  "id": 110223,
  "title": "Full dataset JPEG 512x512",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/110223",
  "author_name": "cab",
  "post_date": "2019-09-26T03:13:00.106000",
  "votes": 41,
  "comment_count": 20,
  "views": 0,
  "content": "<p>I made a full dataset (train + test) JPG with original size 512x512 in <a href=\"https://www.kaggle.com/backaggle/rsna_512\">here</a></p>\n\n<p>The preprocessing code is as follow: \n```\nimport pandas as pd\nimport os\nimport click\nimport glob\nimport cv2\nimport pydicom\nfrom tqdm import tqdm\nfrom utils import get_windowing, window_image\nfrom joblib import delayed, Parallel</p>\n\n<p>@click.group()\ndef cli():\n    print(\"CLI\")</p>\n\n<p>def convert_dicom_to_jpg(dicomfile, outputdir):\n    try:\n        data = pydicom.read_file(dicomfile)\n        image = data.pixel_array\n        window_center, window_width, intercept, slope = get_windowing(data)\n        image_windowed = window_image(image, window_center, window_width, intercept, slope)\n        id = dicomfile.split(\"/\")[-1].split(\".\")[0]\n        output_image = os.path.join(outputdir, id + \".jpg\")\n        cv2.imwrite(output_image, image_windowed)\n    except:\n        print(dicomfile)</p>\n\n<p>@cli.command()\n@click.option('--inputdir', type=str)\n@click.option('--outputdir', type=str)\ndef extract_images(\n    inputdir,\n    outputdir,\n):\n    os.makedirs(outputdir, exist_ok=True)\n    files = glob.glob(inputdir + \"/*.dcm\")\n    Parallel(n_jobs=8)(delayed(convert_dicom_to_jpg)(file, outputdir) for file in tqdm(files, total=len(files)))</p>\n\n<p>if <strong>name</strong> == '<strong>main</strong>':\n    cli()</p>\n\n<p><code>`` \nThe corrupted image is:</code>ID_6431af929<code>\nThe functions (</code>get_windowing<code>,</code>window_image`) are taken from this <a href=\"https://www.kaggle.com/omission/eda-view-dicom-images-with-correct-windowing\">kernel</a>. </p>\n\n<p>Note: \nSaving numpy array to JPG file and reloading the image will lose some pixel information (+-1 value). However, I think it does not really matter. </p>",
  "messages": [
    {
      "id": 634206,
      "postDate": "2019-09-26T03:13:00.107Z",
      "content": "<p>I made a full dataset (train + test) JPG with original size 512x512 in <a href=\"https://www.kaggle.com/backaggle/rsna_512\">here</a></p>\n\n<p>The preprocessing code is as follow: \n```\nimport pandas as pd\nimport os\nimport click\nimport glob\nimport cv2\nimport pydicom\nfrom tqdm import tqdm\nfrom utils import get_windowing, window_image\nfrom joblib import delayed, Parallel</p>\n\n<p>@click.group()\ndef cli():\n    print(\"CLI\")</p>\n\n<p>def convert_dicom_to_jpg(dicomfile, outputdir):\n    try:\n        data = pydicom.read_file(dicomfile)\n        image = data.pixel_array\n        window_center, window_width, intercept, slope = get_windowing(data)\n        image_windowed = window_image(image, window_center, window_width, intercept, slope)\n        id = dicomfile.split(\"/\")[-1].split(\".\")[0]\n        output_image = os.path.join(outputdir, id + \".jpg\")\n        cv2.imwrite(output_image, image_windowed)\n    except:\n        print(dicomfile)</p>\n\n<p>@cli.command()\n@click.option('--inputdir', type=str)\n@click.option('--outputdir', type=str)\ndef extract_images(\n    inputdir,\n    outputdir,\n):\n    os.makedirs(outputdir, exist_ok=True)\n    files = glob.glob(inputdir + \"/*.dcm\")\n    Parallel(n_jobs=8)(delayed(convert_dicom_to_jpg)(file, outputdir) for file in tqdm(files, total=len(files)))</p>\n\n<p>if <strong>name</strong> == '<strong>main</strong>':\n    cli()</p>\n\n<p><code>`` \nThe corrupted image is:</code>ID_6431af929<code>\nThe functions (</code>get_windowing<code>,</code>window_image`) are taken from this <a href=\"https://www.kaggle.com/omission/eda-view-dicom-images-with-correct-windowing\">kernel</a>. </p>\n\n<p>Note: \nSaving numpy array to JPG file and reloading the image will lose some pixel information (+-1 value). However, I think it does not really matter. </p>",
      "rawMarkdown": "I made a full dataset (train + test) JPG with original size 512x512 in [here](https://www.kaggle.com/backaggle/rsna_512)\n\nThe preprocessing code is as follow: \n```\nimport pandas as pd\nimport os\nimport click\nimport glob\nimport cv2\nimport pydicom\nfrom tqdm import tqdm\nfrom utils import get_windowing, window_image\nfrom joblib import delayed, Parallel\n\n\n@click.group()\ndef cli():\n    print(\"CLI\")\n\n\ndef convert_dicom_to_jpg(dicomfile, outputdir):\n    try:\n        data = pydicom.read_file(dicomfile)\n        image = data.pixel_array\n        window_center, window_width, intercept, slope = get_windowing(data)\n        image_windowed = window_image(image, window_center, window_width, intercept, slope)\n        id = dicomfile.split(\"/\")[-1].split(\".\")[0]\n        output_image = os.path.join(outputdir, id + \".jpg\")\n        cv2.imwrite(output_image, image_windowed)\n    except:\n        print(dicomfile)\n\n\n@cli.command()\n@click.option('--inputdir', type=str)\n@click.option('--outputdir', type=str)\ndef extract_images(\n    inputdir,\n    outputdir,\n):\n    os.makedirs(outputdir, exist_ok=True)\n    files = glob.glob(inputdir + \"/*.dcm\")\n    Parallel(n_jobs=8)(delayed(convert_dicom_to_jpg)(file, outputdir) for file in tqdm(files, total=len(files)))\n\n\nif __name__ == '__main__':\n    cli()\n\n``` \nThe corrupted image is: `ID_6431af929`\nThe functions (`get_windowing`,  `window_image`) are taken from this [kernel](https://www.kaggle.com/omission/eda-view-dicom-images-with-correct-windowing). \n\nNote: \nSaving numpy array to JPG file and reloading the image will lose some pixel information (+-1 value). However, I think it does not really matter. ",
      "votes": 41
    },
    {
      "id": 638967,
      "postDate": "2019-10-02T16:15:09.020Z",
      "content": "<p>Do you plan on creating a jpeg data set for stage2? </p>",
      "rawMarkdown": "Do you plan on creating a jpeg data set for stage2? ",
      "votes": 3
    },
    {
      "id": 635851,
      "postDate": "2019-09-28T10:57:04.967Z",
      "content": "<p>Thank you very much for sharing! Appreciate!</p>",
      "rawMarkdown": "Thank you very much for sharing! Appreciate!",
      "votes": 1
    },
    {
      "id": 638116,
      "postDate": "2019-10-01T14:49:05.593Z",
      "content": "<p>I have copied over the PixelData (not the header) from the previous slice (<code>ID_5005bcb25.dcm</code>) to just make it so this \"works\":\n<code>\ndcmdump ID_5005bcb25.dcm +W ./\ncp ID_6431af929.dcm ID_6431af929_new.dcm\ndcmodify -if \"PixelData=ID_5005bcb25.dcm.0.raw\" ID_6431af929_new.dcm\n</code>\nAttached is the updated DICOM.</p>",
      "rawMarkdown": "I have copied over the PixelData (not the header) from the previous slice (`ID_5005bcb25.dcm`) to just make it so this \"works\":\n```\ndcmdump ID_5005bcb25.dcm +W ./\ncp ID_6431af929.dcm ID_6431af929_new.dcm\ndcmodify -if \"PixelData=ID_5005bcb25.dcm.0.raw\" ID_6431af929_new.dcm\n```\nAttached is the updated DICOM.",
      "votes": 2
    },
    {
      "id": 651299,
      "postDate": "2019-10-17T09:49:42.997Z",
      "content": "<p>Thanks for sharing.However some of the images have dimension other than 512x512.</p>",
      "rawMarkdown": "Thanks for sharing.However some of the images have dimension other than 512x512.\n"
    },
    {
      "id": 636232,
      "postDate": "2019-09-29T04:34:13.337Z",
      "content": "<p>Great work! But how did you mark a frame number in filename? Some dicom (maybe most of whole) files are multiframes.</p>",
      "rawMarkdown": "Great work! But how did you mark a frame number in filename? Some dicom (maybe most of whole) files are multiframes."
    },
    {
      "id": 635870,
      "postDate": "2019-09-28T11:39:11.343Z",
      "content": "<p>I am just curious, isn't it that jpg format makes lossy compression, why not png? To make dataset weight lighter?</p>",
      "rawMarkdown": "I am just curious, isn't it that jpg format makes lossy compression, why not png? To make dataset weight lighter?",
      "replies": [
        {
          "id": 635879,
          "postDate": "2019-09-28T11:52:24.283Z",
          "content": "<p>Because I use jpeg4py which can read jpg faster than png. As usual, I prefer jpg rather than other formats.</p>",
          "rawMarkdown": "Because I use jpeg4py which can read jpg faster than png. As usual, I prefer jpg rather than other formats.",
          "votes": 1
        }
      ]
    },
    {
      "id": 635736,
      "postDate": "2019-09-28T05:56:31.187Z",
      "content": "<p>how did you upload on kaggle.\nI tried to uupload from Colab after processing using \"kaggle datasets .....\" command it get uploaded and once I got link coming after successful run of command got 404 error .\nAny idea .\nThanks.</p>",
      "rawMarkdown": "how did you upload on kaggle.\nI tried to uupload from Colab after processing using \"kaggle datasets .....\" command it get uploaded and once I got link coming after successful run of command got 404 error .\nAny idea .\nThanks.",
      "replies": [
        {
          "id": 635747,
          "postDate": "2019-09-28T06:12:04.407Z",
          "content": "<p>If you want to upload &gt;20G dataset, you should make it public. Make sure you set: 'isPrivate: False' in the metadata file init. </p>",
          "rawMarkdown": "If you want to upload &gt;20G dataset, you should make it public. Make sure you set: 'isPrivate: False' in the metadata file init. "
        }
      ]
    },
    {
      "id": 635700,
      "postDate": "2019-09-28T04:12:43.240Z",
      "content": "<p><a href=\"/backaggle\">@backaggle</a> what kind of compression rate did you use when you zipped your images - standard, high? Thanks.</p>",
      "rawMarkdown": "@backaggle what kind of compression rate did you use when you zipped your images - standard, high? Thanks.",
      "replies": [
        {
          "id": 635746,
          "postDate": "2019-09-28T06:10:09.940Z",
          "content": "<p>I use Kaggle API to update the dataset. It automatically compresses folder itself. I think it uses tar's default settings</p>",
          "rawMarkdown": "I use Kaggle API to update the dataset. It automatically compresses folder itself. I think it uses tar's default settings",
          "votes": 1
        },
        {
          "id": 636184,
          "postDate": "2019-09-29T00:33:12.017Z",
          "content": "<p><a href=\"/backaggle\">@backaggle</a> How long did it take you to upload your dataset using the kaggle API? I've tried to zip and upload my own images and a smaller resolution using 7zip and it takes forever to upload (6GB). When downloading the dataset into kernel it also takes forever. What is the API command you used to upload your dataset? Help would be appreciated!</p>",
          "rawMarkdown": "@backaggle How long did it take you to upload your dataset using the kaggle API? I've tried to zip and upload my own images and a smaller resolution using 7zip and it takes forever to upload (6GB). When downloading the dataset into kernel it also takes forever. What is the API command you used to upload your dataset? Help would be appreciated!"
        },
        {
          "id": 636205,
          "postDate": "2019-09-29T02:56:29.137Z",
          "content": "<ol>\n<li>You should create dataset metadata \n<code>kaggle datasets init</code> </li>\n</ol>\n\n<p>Add <code>isPrivate: false</code> to <code>dataset-metadata.json</code> to make it public if you want. Default is private. </p>\n\n<ol>\n<li>Upload dataset \nEverything in the same folder as <code>dataset-metadata.json</code> is considered as the data. To update them, \n<code>kaggle datasets create -r</code> </li>\n</ol>",
          "rawMarkdown": "1. You should create dataset metadata \n`kaggle datasets init` \n\nAdd `isPrivate: false` to `dataset-metadata.json` to make it public if you want. Default is private. \n\n2. Upload dataset \nEverything in the same folder as `dataset-metadata.json` is considered as the data. To update them, \n`kaggle datasets create -r` ",
          "votes": 1
        },
        {
          "id": 636726,
          "postDate": "2019-09-30T05:26:32.653Z",
          "content": "<p>in my json file i set<code>\"path\": \"E:/test/\",</code> where do I need to put the dataset-metadata.json file? when I run the <code>kaggle datasets create -p [filepath]</code> what do I put as the filepath? I can't figure this out... I am basically confused on three parts here: json path, where to put the json file and what the filepath should be in the kaggle api command.</p>",
          "rawMarkdown": "in my json file i set`\"path\": \"E:/test/\",` where do I need to put the dataset-metadata.json file? when I run the `kaggle datasets create -p [filepath]` what do I put as the filepath? I can't figure this out... I am basically confused on three parts here: json path, where to put the json file and what the filepath should be in the kaggle api command."
        }
      ]
    },
    {
      "id": 635506,
      "postDate": "2019-09-27T18:03:54.297Z",
      "content": "<p>Hi Bac, thanks for the dataset.\nI'm trying to use it, but can't add to the kernel.\nI don't know why.\nI can use the othes datasets of the others fellows: png 128, 224, etc, but can't add yours.\nTell me if you figure out why it is happening, please.</p>",
      "rawMarkdown": "Hi Bac, thanks for the dataset.\nI'm trying to use it, but can't add to the kernel.\nI don't know why.\nI can use the othes datasets of the others fellows: png 128, 224, etc, but can't add yours.\nTell me if you figure out why it is happening, please."
    },
    {
      "id": 635740,
      "postDate": "2019-09-28T05:58:48.743Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 635316,
      "postDate": "2019-09-27T11:24:48.677Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 640257,
      "postDate": "2019-10-04T00:20:16.733Z",
      "content": "<p>Thanks for your sharing!</p>",
      "rawMarkdown": "Thanks for your sharing!",
      "votes": 1
    },
    {
      "id": 635743,
      "postDate": "2019-09-28T06:04:42.337Z",
      "content": "<p>Thanks for your dataset.</p>",
      "rawMarkdown": "Thanks for your dataset.",
      "votes": 1
    },
    {
      "id": 651203,
      "postDate": "2019-10-17T07:10:49.283Z",
      "content": "<p>Thanks for your sharing!</p>",
      "rawMarkdown": "Thanks for your sharing!"
    }
  ],
  "comments": [
    {
      "id": 638967,
      "author_name": "William Green",
      "author_url": "",
      "post_date": "2019-10-02T16:15:09.020000",
      "content": "<p>Do you plan on creating a jpeg data set for stage2? </p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 635851,
      "author_name": "Rui Lebre",
      "author_url": "",
      "post_date": "2019-09-28T10:57:04.967000",
      "content": "<p>Thank you very much for sharing! Appreciate!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 638116,
      "author_name": "John M",
      "author_url": "",
      "post_date": "2019-10-01T14:49:05.593000",
      "content": "<p>I have copied over the PixelData (not the header) from the previous slice (<code>ID_5005bcb25.dcm</code>) to just make it so this \"works\":\n<code>\ndcmdump ID_5005bcb25.dcm +W ./\ncp ID_6431af929.dcm ID_6431af929_new.dcm\ndcmodify -if \"PixelData=ID_5005bcb25.dcm.0.raw\" ID_6431af929_new.dcm\n</code>\nAttached is the updated DICOM.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 651299,
      "author_name": "Deepshad",
      "author_url": "",
      "post_date": "2019-10-17T09:49:42.997000",
      "content": "<p>Thanks for sharing.However some of the images have dimension other than 512x512.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 636232,
      "author_name": "Alexander Fedyukov",
      "author_url": "",
      "post_date": "2019-09-29T04:34:13.337000",
      "content": "<p>Great work! But how did you mark a frame number in filename? Some dicom (maybe most of whole) files are multiframes.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 635870,
      "author_name": "Kuda",
      "author_url": "",
      "post_date": "2019-09-28T11:39:11.343000",
      "content": "<p>I am just curious, isn't it that jpg format makes lossy compression, why not png? To make dataset weight lighter?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 635879,
          "author_name": "cab",
          "author_url": "",
          "post_date": "2019-09-28T11:52:24.283000",
          "content": "<p>Because I use jpeg4py which can read jpg faster than png. As usual, I prefer jpg rather than other formats.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 635736,
      "author_name": "Rajnish Chauhan",
      "author_url": "",
      "post_date": "2019-09-28T05:56:31.187000",
      "content": "<p>how did you upload on kaggle.\nI tried to uupload from Colab after processing using \"kaggle datasets .....\" command it get uploaded and once I got link coming after successful run of command got 404 error .\nAny idea .\nThanks.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 635747,
          "author_name": "cab",
          "author_url": "",
          "post_date": "2019-09-28T06:12:04.407000",
          "content": "<p>If you want to upload &gt;20G dataset, you should make it public. Make sure you set: 'isPrivate: False' in the metadata file init. </p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 635700,
      "author_name": "Tim Yee",
      "author_url": "",
      "post_date": "2019-09-28T04:12:43.240000",
      "content": "<p><a href=\"/backaggle\">@backaggle</a> what kind of compression rate did you use when you zipped your images - standard, high? Thanks.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 635746,
          "author_name": "cab",
          "author_url": "",
          "post_date": "2019-09-28T06:10:09.940000",
          "content": "<p>I use Kaggle API to update the dataset. It automatically compresses folder itself. I think it uses tar's default settings</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 636184,
          "author_name": "Tim Yee",
          "author_url": "",
          "post_date": "2019-09-29T00:33:12.017000",
          "content": "<p><a href=\"/backaggle\">@backaggle</a> How long did it take you to upload your dataset using the kaggle API? I've tried to zip and upload my own images and a smaller resolution using 7zip and it takes forever to upload (6GB). When downloading the dataset into kernel it also takes forever. What is the API command you used to upload your dataset? Help would be appreciated!</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 636205,
          "author_name": "cab",
          "author_url": "",
          "post_date": "2019-09-29T02:56:29.137000",
          "content": "<ol>\n<li>You should create dataset metadata \n<code>kaggle datasets init</code> </li>\n</ol>\n\n<p>Add <code>isPrivate: false</code> to <code>dataset-metadata.json</code> to make it public if you want. Default is private. </p>\n\n<ol>\n<li>Upload dataset \nEverything in the same folder as <code>dataset-metadata.json</code> is considered as the data. To update them, \n<code>kaggle datasets create -r</code> </li>\n</ol>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 636726,
          "author_name": "Tim Yee",
          "author_url": "",
          "post_date": "2019-09-30T05:26:32.653000",
          "content": "<p>in my json file i set<code>\"path\": \"E:/test/\",</code> where do I need to put the dataset-metadata.json file? when I run the <code>kaggle datasets create -p [filepath]</code> what do I put as the filepath? I can't figure this out... I am basically confused on three parts here: json path, where to put the json file and what the filepath should be in the kaggle api command.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 635506,
      "author_name": "Bruno Aquino",
      "author_url": "",
      "post_date": "2019-09-27T18:03:54.297000",
      "content": "<p>Hi Bac, thanks for the dataset.\nI'm trying to use it, but can't add to the kernel.\nI don't know why.\nI can use the othes datasets of the others fellows: png 128, 224, etc, but can't add yours.\nTell me if you figure out why it is happening, please.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 635740,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-09-28T05:58:48.743000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 635316,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-09-27T11:24:48.677000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 640257,
      "author_name": "sw_tju",
      "author_url": "",
      "post_date": "2019-10-04T00:20:16.733000",
      "content": "<p>Thanks for your sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 635743,
      "author_name": "Quang Trinh",
      "author_url": "",
      "post_date": "2019-09-28T06:04:42.337000",
      "content": "<p>Thanks for your dataset.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 651203,
      "author_name": "LongYin/杰少",
      "author_url": "",
      "post_date": "2019-10-17T07:10:49.283000",
      "content": "<p>Thanks for your sharing!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "634206": "I made a full dataset (train + test) JPG with original size 512x512 in [here](https://www.kaggle.com/backaggle/rsna_512)\n\nThe preprocessing code is as follow: \n```\nimport pandas as pd\nimport os\nimport click\nimport glob\nimport cv2\nimport pydicom\nfrom tqdm import tqdm\nfrom utils import get_windowing, window_image\nfrom joblib import delayed, Parallel\n\n\n@click.group()\ndef cli():\n    print(\"CLI\")\n\n\ndef convert_dicom_to_jpg(dicomfile, outputdir):\n    try:\n        data = pydicom.read_file(dicomfile)\n        image = data.pixel_array\n        window_center, window_width, intercept, slope = get_windowing(data)\n        image_windowed = window_image(image, window_center, window_width, intercept, slope)\n        id = dicomfile.split(\"/\")[-1].split(\".\")[0]\n        output_image = os.path.join(outputdir, id + \".jpg\")\n        cv2.imwrite(output_image, image_windowed)\n    except:\n        print(dicomfile)\n\n\n@cli.command()\n@click.option('--inputdir', type=str)\n@click.option('--outputdir', type=str)\ndef extract_images(\n    inputdir,\n    outputdir,\n):\n    os.makedirs(outputdir, exist_ok=True)\n    files = glob.glob(inputdir + \"/*.dcm\")\n    Parallel(n_jobs=8)(delayed(convert_dicom_to_jpg)(file, outputdir) for file in tqdm(files, total=len(files)))\n\n\nif __name__ == '__main__':\n    cli()\n\n``` \nThe corrupted image is: `ID_6431af929`\nThe functions (`get_windowing`,  `window_image`) are taken from this [kernel](https://www.kaggle.com/omission/eda-view-dicom-images-with-correct-windowing). \n\nNote: \nSaving numpy array to JPG file and reloading the image will lose some pixel information (+-1 value). However, I think it does not really matter. ",
    "638967": "Do you plan on creating a jpeg data set for stage2? ",
    "635851": "Thank you very much for sharing! Appreciate!",
    "638116": "I have copied over the PixelData (not the header) from the previous slice (`ID_5005bcb25.dcm`) to just make it so this \"works\":\n```\ndcmdump ID_5005bcb25.dcm +W ./\ncp ID_6431af929.dcm ID_6431af929_new.dcm\ndcmodify -if \"PixelData=ID_5005bcb25.dcm.0.raw\" ID_6431af929_new.dcm\n```\nAttached is the updated DICOM.",
    "651299": "Thanks for sharing.However some of the images have dimension other than 512x512.\n",
    "636232": "Great work! But how did you mark a frame number in filename? Some dicom (maybe most of whole) files are multiframes.",
    "635870": "I am just curious, isn't it that jpg format makes lossy compression, why not png? To make dataset weight lighter?",
    "635736": "how did you upload on kaggle.\nI tried to uupload from Colab after processing using \"kaggle datasets .....\" command it get uploaded and once I got link coming after successful run of command got 404 error .\nAny idea .\nThanks.",
    "635700": "@backaggle what kind of compression rate did you use when you zipped your images - standard, high? Thanks.",
    "635506": "Hi Bac, thanks for the dataset.\nI'm trying to use it, but can't add to the kernel.\nI don't know why.\nI can use the othes datasets of the others fellows: png 128, 224, etc, but can't add yours.\nTell me if you figure out why it is happening, please.",
    "635740": "",
    "635316": "",
    "640257": "Thanks for your sharing!",
    "635743": "Thanks for your dataset.",
    "651203": "Thanks for your sharing!"
  }
}