{
  "id": 20051,
  "title": "Beginning the challenge: Manually label about 20,000 images?!",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/20051",
  "author_name": "",
  "post_date": "2016-04-10T18:55:25.783Z",
  "votes": null,
  "comment_count": 3,
  "views": 1435,
  "content": "<p>Hi,</p>\n\n<p>I am about to begin working on this challenge, but I don't understand how (and why) should I label the training images manually? Isn't the point of a machine learning problem to enable the machine to predict the labels, rather than the humans doing it? \nIn case we do have to label it, where do I begin?</p>\n\n<p>Please let me know what you guys think. Cheers.</p>",
  "messages": [
    {
      "id": "114429",
      "postDate": "04/10/2016 18:55:25",
      "content": "<p>Hi,</p>\n\n<p>I am about to begin working on this challenge, but I don't understand how (and why) should I label the training images manually? Isn't the point of a machine learning problem to enable the machine to predict the labels, rather than the humans doing it? \nIn case we do have to label it, where do I begin?</p>\n\n<p>Please let me know what you guys think. Cheers.</p>",
      "rawMarkdown": "Hi,\r\n\r\nI am about to begin working on this challenge, but I don't understand how (and why) should I label the training images manually? Isn't the point of a machine learning problem to enable the machine to predict the labels, rather than the humans doing it? \r\nIn case we do have to label it, where do I begin?\r\n\r\nPlease let me know what you guys think. Cheers.",
      "votes": null
    },
    {
      "id": "114432",
      "postDate": "04/10/2016 18:58:42",
      "content": "<p>Folders are the labels</p>",
      "rawMarkdown": "Folders are the labels",
      "votes": null
    },
    {
      "id": "114436",
      "postDate": "04/10/2016 20:33:21",
      "content": "<p>If you've seen any discussion about manually labeling the images, it is not about labeling the class (which, as @Roman notes, is given by the folder). Rather, it is about labeling features that may be useful for training a classifier (e.g., the location of the driver's head in the image).</p>",
      "rawMarkdown": "If you've seen any discussion about manually labeling the images, it is not about labeling the class (which, as @Roman notes, is given by the folder). Rather, it is about labeling features that may be useful for training a classifier (e.g., the location of the driver's head in the image).",
      "votes": null
    },
    {
      "id": "114521",
      "postDate": "04/11/2016 15:52:08",
      "content": "<p>Hi Abhijay,</p>\n\n<p>Please refer to code in TensorFlow where the images are labeled based on  folder names</p>\n\n<p><a href=\"https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb\">https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb</a></p>",
      "rawMarkdown": "Hi Abhijay,\r\n\r\nPlease refer to code in TensorFlow where the images are labeled based on  folder names\r\n\r\nhttps://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 114432,
      "author_name": "inoryy",
      "author_url": "",
      "post_date": "04/10/2016 18:58:42",
      "content": "<p>Folders are the labels</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 114436,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "04/10/2016 20:33:21",
      "content": "<p>If you've seen any discussion about manually labeling the images, it is not about labeling the class (which, as @Roman notes, is given by the folder). Rather, it is about labeling features that may be useful for training a classifier (e.g., the location of the driver's head in the image).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 114521,
      "author_name": "vdpappu",
      "author_url": "",
      "post_date": "04/11/2016 15:52:08",
      "content": "<p>Hi Abhijay,</p>\n\n<p>Please refer to code in TensorFlow where the images are labeled based on  folder names</p>\n\n<p><a href=\"https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb\">https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "114429": "Hi,\r\n\r\nI am about to begin working on this challenge, but I don't understand how (and why) should I label the training images manually? Isn't the point of a machine learning problem to enable the machine to predict the labels, rather than the humans doing it? \r\nIn case we do have to label it, where do I begin?\r\n\r\nPlease let me know what you guys think. Cheers.",
    "114432": "Folders are the labels",
    "114436": "If you've seen any discussion about manually labeling the images, it is not about labeling the class (which, as @Roman notes, is given by the folder). Rather, it is about labeling features that may be useful for training a classifier (e.g., the location of the driver's head in the image).",
    "114521": "Hi Abhijay,\r\n\r\nPlease refer to code in TensorFlow where the images are labeled based on  folder names\r\n\r\nhttps://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/1_notmnist.ipynb"
  },
  "source": "meta"
}