{
  "id": 290942,
  "title": "Understanding Data for Petals to Metal",
  "url": "/competitions/tpu-getting-started/discussion/290942",
  "author_name": "Manish Gupta",
  "post_date": "2021-11-26T22:38:40.284000",
  "votes": 4,
  "comment_count": 1,
  "views": null,
  "content": "<p><a href=\"https://www.kaggle.com/alexisbcook\" target=\"_blank\">@alexisbcook</a> <a href=\"https://www.kaggle.com/ryanholbrook\" target=\"_blank\">@ryanholbrook</a> <br>\nI am trying to learn Deep Learning and Computer Vision from wonderful resources available on kaggle. However I have stuck up at the very beginning. I can see from Data Explorer section of Petals to Metal competition page that there are 4 main folders with each one having again 3 subfolders train, val, test which contain acutal data. The 4 mian folders are:<br>\ntfrecords-jpeg-512x512<br>\ntfrecords-jpeg-331x331<br>\ntfrecords-jpeg-224x224<br>\ntfrecords-jpeg-192x192<br>\nIt seems that these each of these 4 folders contains the the same train, test, val data for images but with different pixel values since no. of images are exactly the same in each .tfrec file in each of these 4 main folders.<br>\n![<a href=\"https://ibb.co/fY16t07\" target=\"_blank\">https://ibb.co/fY16t07</a>]<br>\nIf my understanding is correct that can someone please explain why the same images with different pixel values are provided and whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?<br>\nI am newbie and any help will be highly appreciated.</p>",
  "messages": [
    {
      "id": 2203063,
      "postDate": "2023-03-30T13:40:31.430Z",
      "content": "<h1>1:  why the same images with different pixel values are provided?</h1>\n<p>You may choose the pixel variants which you find most suitable. For instance, using 512 might not always give the best results. It will definitely require more infra </p>\n<h1>2: whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?</h1>\n<p>Yes, you should use the same resolution for model train, val, test</p>",
      "rawMarkdown": "#1:  why the same images with different pixel values are provided?\nYou may choose the pixel variants which you find most suitable. For instance, using 512 might not always give the best results. It will definitely require more infra \n\n\n#2: whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?\nYes, you should use the same resolution for model train, val, test\n",
      "votes": 1
    },
    {
      "id": 1596842,
      "postDate": "2021-11-26T22:38:40.283Z",
      "content": "<p><a href=\"https://www.kaggle.com/alexisbcook\" target=\"_blank\">@alexisbcook</a> <a href=\"https://www.kaggle.com/ryanholbrook\" target=\"_blank\">@ryanholbrook</a> <br>\nI am trying to learn Deep Learning and Computer Vision from wonderful resources available on kaggle. However I have stuck up at the very beginning. I can see from Data Explorer section of Petals to Metal competition page that there are 4 main folders with each one having again 3 subfolders train, val, test which contain acutal data. The 4 mian folders are:<br>\ntfrecords-jpeg-512x512<br>\ntfrecords-jpeg-331x331<br>\ntfrecords-jpeg-224x224<br>\ntfrecords-jpeg-192x192<br>\nIt seems that these each of these 4 folders contains the the same train, test, val data for images but with different pixel values since no. of images are exactly the same in each .tfrec file in each of these 4 main folders.<br>\n![<a href=\"https://ibb.co/fY16t07\" target=\"_blank\">https://ibb.co/fY16t07</a>]<br>\nIf my understanding is correct that can someone please explain why the same images with different pixel values are provided and whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?<br>\nI am newbie and any help will be highly appreciated.</p>",
      "rawMarkdown": "@alexisbcook @ryanholbrook \nI am trying to learn Deep Learning and Computer Vision from wonderful resources available on kaggle. However I have stuck up at the very beginning. I can see from Data Explorer section of Petals to Metal competition page that there are 4 main folders with each one having again 3 subfolders train, val, test which contain acutal data. The 4 mian folders are:\ntfrecords-jpeg-512x512\ntfrecords-jpeg-331x331\ntfrecords-jpeg-224x224\ntfrecords-jpeg-192x192\nIt seems that these each of these 4 folders contains the the same train, test, val data for images but with different pixel values since no. of images are exactly the same in each .tfrec file in each of these 4 main folders.\n![https://ibb.co/fY16t07]\nIf my understanding is correct that can someone please explain why the same images with different pixel values are provided and whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?\nI am newbie and any help will be highly appreciated.",
      "votes": 4
    }
  ],
  "comments": [
    {
      "id": 2203063,
      "author_name": "RahulBhambri",
      "author_url": "",
      "post_date": "2023-03-30T13:40:31.430000",
      "content": "<h1>1:  why the same images with different pixel values are provided?</h1>\n<p>You may choose the pixel variants which you find most suitable. For instance, using 512 might not always give the best results. It will definitely require more infra </p>\n<h1>2: whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?</h1>\n<p>Yes, you should use the same resolution for model train, val, test</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2203063": "#1:  why the same images with different pixel values are provided?\nYou may choose the pixel variants which you find most suitable. For instance, using 512 might not always give the best results. It will definitely require more infra \n\n\n#2: whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?\nYes, you should use the same resolution for model train, val, test\n",
    "1596842": "@alexisbcook @ryanholbrook \nI am trying to learn Deep Learning and Computer Vision from wonderful resources available on kaggle. However I have stuck up at the very beginning. I can see from Data Explorer section of Petals to Metal competition page that there are 4 main folders with each one having again 3 subfolders train, val, test which contain acutal data. The 4 mian folders are:\ntfrecords-jpeg-512x512\ntfrecords-jpeg-331x331\ntfrecords-jpeg-224x224\ntfrecords-jpeg-192x192\nIt seems that these each of these 4 folders contains the the same train, test, val data for images but with different pixel values since no. of images are exactly the same in each .tfrec file in each of these 4 main folders.\n![https://ibb.co/fY16t07]\nIf my understanding is correct that can someone please explain why the same images with different pixel values are provided and whether we need to use image data only in any one main folder, say tfrecords-jpeg-512x512 for train, val and test?\nI am newbie and any help will be highly appreciated."
  }
}