{
  "id": 15912,
  "title": "Using Overfeat as classifier: general categories?",
  "url": "/competitions/dogs-vs-cats/discussion/15912",
  "author_name": "",
  "post_date": "2015-08-12T17:49:00.950Z",
  "votes": 2,
  "comment_count": 2,
  "views": 1432,
  "content": "<p>Hello.</p>\n\n<p>I am trying to use Overfeat as classifier with just the pre-trained ImageNet data, to see how good it performs for this challenge.</p>\n\n<p>The problem is the classifier extracts very specific classes such as &quot;fox terrier&quot;, &quot;dalmatian&quot;, &quot;persian cat&quot;, making it difficult to automatically assign each result to &quot;dog&quot; or &quot;cat&quot;.</p>\n\n<p>Is there any way to extract more general classes? It looks like ImageNet has subtrees for &quot;domestic cat&quot; and &quot;domestic dog&quot;, but I am not sure if those are used as classes or just for visual organization.</p>\n\n<p>Thank you very much.</p>",
  "messages": [
    {
      "id": "89201",
      "postDate": "08/12/2015 17:49:00",
      "content": "<p>Hello.</p>\n\n<p>I am trying to use Overfeat as classifier with just the pre-trained ImageNet data, to see how good it performs for this challenge.</p>\n\n<p>The problem is the classifier extracts very specific classes such as &quot;fox terrier&quot;, &quot;dalmatian&quot;, &quot;persian cat&quot;, making it difficult to automatically assign each result to &quot;dog&quot; or &quot;cat&quot;.</p>\n\n<p>Is there any way to extract more general classes? It looks like ImageNet has subtrees for &quot;domestic cat&quot; and &quot;domestic dog&quot;, but I am not sure if those are used as classes or just for visual organization.</p>\n\n<p>Thank you very much.</p>",
      "rawMarkdown": "Hello.\r\n\r\nI am trying to use Overfeat as classifier with just the pre-trained ImageNet data, to see how good it performs for this challenge.\r\n\r\nThe problem is the classifier extracts very specific classes such as \"fox terrier\", \"dalmatian\", \"persian cat\", making it difficult to automatically assign each result to \"dog\" or \"cat\".\r\n\r\nIs there any way to extract more general classes? It looks like ImageNet has subtrees for \"domestic cat\" and \"domestic dog\", but I am not sure if those are used as classes or just for visual organization.\r\n\r\nThank you very much.",
      "votes": null
    },
    {
      "id": "90560",
      "postDate": "08/27/2015 20:07:51",
      "content": "<p>That would be awesome! Any news @DavidPerra ?</p>",
      "rawMarkdown": "That would be awesome! Any news @DavidPerra ?",
      "votes": null
    },
    {
      "id": "90563",
      "postDate": "08/27/2015 20:24:04",
      "content": "<p>I learned that is better to use Overfeat as feature extractor and then use these features to train a 2-class classifier.</p>",
      "rawMarkdown": "I learned that is better to use Overfeat as feature extractor and then use these features to train a 2-class classifier.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 90560,
      "author_name": "wikunia",
      "author_url": "",
      "post_date": "08/27/2015 20:07:51",
      "content": "<p>That would be awesome! Any news @DavidPerra ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 90563,
      "author_name": "karmapolice",
      "author_url": "",
      "post_date": "08/27/2015 20:24:04",
      "content": "<p>I learned that is better to use Overfeat as feature extractor and then use these features to train a 2-class classifier.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "89201": "Hello.\r\n\r\nI am trying to use Overfeat as classifier with just the pre-trained ImageNet data, to see how good it performs for this challenge.\r\n\r\nThe problem is the classifier extracts very specific classes such as \"fox terrier\", \"dalmatian\", \"persian cat\", making it difficult to automatically assign each result to \"dog\" or \"cat\".\r\n\r\nIs there any way to extract more general classes? It looks like ImageNet has subtrees for \"domestic cat\" and \"domestic dog\", but I am not sure if those are used as classes or just for visual organization.\r\n\r\nThank you very much.",
    "90560": "That would be awesome! Any news @DavidPerra ?",
    "90563": "I learned that is better to use Overfeat as feature extractor and then use these features to train a 2-class classifier."
  },
  "source": "meta"
}