{
  "id": 48105,
  "title": "What dataset are you using?",
  "url": "/competitions/sp-society-camera-model-identification/discussion/48105",
  "author_name": "",
  "post_date": "2018-01-23T14:53:54.287191Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I'm currently using Gleb's dataset (<a href=\"https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235\">https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235</a>)  plus the organization one. </p>\n\n<p>Which one are you using?</p>",
  "messages": [
    {
      "id": "272710",
      "postDate": "01/23/2018 14:53:54",
      "content": "<p>I'm currently using Gleb's dataset (<a href=\"https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235\">https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235</a>)  plus the organization one. </p>\n\n<p>Which one are you using?</p>",
      "rawMarkdown": "I'm currently using Gleb's dataset (https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235)  plus the organization one. \n\nWhich one are you using?",
      "votes": null
    },
    {
      "id": "272711",
      "postDate": "01/23/2018 15:03:06",
      "content": "<p>The same, with the Gleb validation files for validation. I read the whole image into memory and take a random 224 x 224 crop which is passed to Resnet with the first 50 or so layers having layer.trainable set to false. Since I'm doing one image at a time the training is pretty slow, I need to figure out how to get a whole batch loaded into memory efficiently.</p>\n\n<p>A big advantage of using the Gleb validation files is that my validation score is pretty close to my public LB score.</p>",
      "rawMarkdown": "The same, with the Gleb validation files for validation. I read the whole image into memory and take a random 224 x 224 crop which is passed to Resnet with the first 50 or so layers having layer.trainable set to false. Since I'm doing one image at a time the training is pretty slow, I need to figure out how to get a whole batch loaded into memory efficiently.\n\nA big advantage of using the Gleb validation files is that my validation score is pretty close to my public LB score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 272711,
      "author_name": "jfkingiii",
      "author_url": "",
      "post_date": "01/23/2018 15:03:06",
      "content": "<p>The same, with the Gleb validation files for validation. I read the whole image into memory and take a random 224 x 224 crop which is passed to Resnet with the first 50 or so layers having layer.trainable set to false. Since I'm doing one image at a time the training is pretty slow, I need to figure out how to get a whole batch loaded into memory efficiently.</p>\n\n<p>A big advantage of using the Gleb validation files is that my validation score is pretty close to my public LB score.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "272710": "I'm currently using Gleb's dataset (https://www.kaggle.com/c/sp-society-camera-model-identification/discussion/47235)  plus the organization one. \n\nWhich one are you using?",
    "272711": "The same, with the Gleb validation files for validation. I read the whole image into memory and take a random 224 x 224 crop which is passed to Resnet with the first 50 or so layers having layer.trainable set to false. Since I'm doing one image at a time the training is pretty slow, I need to figure out how to get a whole batch loaded into memory efficiently.\n\nA big advantage of using the Gleb validation files is that my validation score is pretty close to my public LB score."
  },
  "source": "meta"
}