{
  "id": 33489,
  "title": "Image preprocessing/resizing",
  "url": "/competitions/noaa-fisheries-steller-sea-lion-population-count/discussion/33489",
  "author_name": "",
  "post_date": "2017-05-24T14:20:23.641489200Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I want to learn how to run CNNs on this dataset. Do I need to resize these pictures? Maybe 200x200ish for me to run these networks? Would that be too small, given that the sea lion's are really small in these pictures?</p>",
  "messages": [
    {
      "id": "185217",
      "postDate": "05/24/2017 14:20:23",
      "content": "<p>I want to learn how to run CNNs on this dataset. Do I need to resize these pictures? Maybe 200x200ish for me to run these networks? Would that be too small, given that the sea lion's are really small in these pictures?</p>",
      "rawMarkdown": "I want to learn how to run CNNs on this dataset. Do I need to resize these pictures? Maybe 200x200ish for me to run these networks? Would that be too small, given that the sea lion's are really small in these pictures?",
      "votes": null
    },
    {
      "id": "185317",
      "postDate": "05/24/2017 21:15:49",
      "content": "<p>It will be small. Generally, people cut image into patches. For example 150x150 px and the train a network on all the patches. Then You preserve the original scale.</p>",
      "rawMarkdown": "It will be small. Generally, people cut image into patches. For example 150x150 px and the train a network on all the patches. Then You preserve the original scale.",
      "votes": null
    },
    {
      "id": "185365",
      "postDate": "05/25/2017 00:23:25",
      "content": "<p>Would it be easy to recreate your y labels based on the cuts you made?</p>",
      "rawMarkdown": "Would it be easy to recreate your y labels based on the cuts you made?",
      "votes": null
    },
    {
      "id": "185568",
      "postDate": "05/25/2017 12:16:22",
      "content": "<p>it depends on your format of y labels. For sea lions dataset, since you know positions of every sea lion, you may generate the correct y labels for the crop. May be tricky to handle lions on edges of crop.</p>",
      "rawMarkdown": "it depends on your format of y labels. For sea lions dataset, since you know positions of every sea lion, you may generate the correct y labels for the crop. May be tricky to handle lions on edges of crop.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 185317,
      "author_name": "asanakoev",
      "author_url": "",
      "post_date": "05/24/2017 21:15:49",
      "content": "<p>It will be small. Generally, people cut image into patches. For example 150x150 px and the train a network on all the patches. Then You preserve the original scale.</p>",
      "votes": null,
      "replies": [
        {
          "id": 185365,
          "author_name": "terranceliu",
          "author_url": "",
          "post_date": "05/25/2017 00:23:25",
          "content": "<p>Would it be easy to recreate your y labels based on the cuts you made?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 185568,
          "author_name": "dmytropoplavskiy",
          "author_url": "",
          "post_date": "05/25/2017 12:16:22",
          "content": "<p>it depends on your format of y labels. For sea lions dataset, since you know positions of every sea lion, you may generate the correct y labels for the crop. May be tricky to handle lions on edges of crop.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "185217": "I want to learn how to run CNNs on this dataset. Do I need to resize these pictures? Maybe 200x200ish for me to run these networks? Would that be too small, given that the sea lion's are really small in these pictures?",
    "185317": "It will be small. Generally, people cut image into patches. For example 150x150 px and the train a network on all the patches. Then You preserve the original scale.",
    "185365": "Would it be easy to recreate your y labels based on the cuts you made?",
    "185568": "it depends on your format of y labels. For sea lions dataset, since you know positions of every sea lion, you may generate the correct y labels for the crop. May be tricky to handle lions on edges of crop."
  },
  "source": "meta"
}