{
  "id": 60792,
  "title": "Haar cascades",
  "url": "/competitions/google-ai-open-images-object-detection-track/discussion/60792",
  "author_name": "Rémi",
  "post_date": "2018-07-10T02:11:33.774000",
  "votes": -2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Are haar cascades even relevant for that type of learning</p>",
  "messages": [
    {
      "id": 354659,
      "postDate": "2018-07-10T02:11:33.773Z",
      "content": "<p>Are haar cascades even relevant for that type of learning</p>",
      "rawMarkdown": "Are haar cascades even relevant for that type of learning",
      "votes": -2
    },
    {
      "id": 355321,
      "postDate": "2018-07-11T13:05:45.717Z",
      "content": "<p>My experience with HAAR cascades is that it is useful for a very specific template style detection, like frontal faces. It's extremely fast, but doesn't seem to have any capability to learn higher level abstractions of image features, like a deep neural network can.</p>\n\n<p>If I'm honest, I do not think HAAR cascades will get anywhere in this type of challenge. I think the place to start are some popular object detection networks, such as YOLO, SSD, or RCNN.</p>",
      "rawMarkdown": "My experience with HAAR cascades is that it is useful for a very specific template style detection, like frontal faces. It's extremely fast, but doesn't seem to have any capability to learn higher level abstractions of image features, like a deep neural network can.\n\nIf I'm honest, I do not think HAAR cascades will get anywhere in this type of challenge. I think the place to start are some popular object detection networks, such as YOLO, SSD, or RCNN.",
      "votes": 2,
      "isDeleted": true,
      "replies": [
        {
          "id": 359307,
          "postDate": "2018-07-19T20:58:50.743Z",
          "content": "<p>Thank you very much</p>",
          "rawMarkdown": "Thank you very much"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 355321,
      "author_name": "",
      "author_url": "",
      "post_date": "2018-07-11T13:05:45.717000",
      "content": "<p>My experience with HAAR cascades is that it is useful for a very specific template style detection, like frontal faces. It's extremely fast, but doesn't seem to have any capability to learn higher level abstractions of image features, like a deep neural network can.</p>\n\n<p>If I'm honest, I do not think HAAR cascades will get anywhere in this type of challenge. I think the place to start are some popular object detection networks, such as YOLO, SSD, or RCNN.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 359307,
          "author_name": "Rémi",
          "author_url": "",
          "post_date": "2018-07-19T20:58:50.743000",
          "content": "<p>Thank you very much</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "354659": "Are haar cascades even relevant for that type of learning",
    "355321": "My experience with HAAR cascades is that it is useful for a very specific template style detection, like frontal faces. It's extremely fast, but doesn't seem to have any capability to learn higher level abstractions of image features, like a deep neural network can.\n\nIf I'm honest, I do not think HAAR cascades will get anywhere in this type of challenge. I think the place to start are some popular object detection networks, such as YOLO, SSD, or RCNN."
  }
}