{
  "id": 64856,
  "title": "Model Advice: YOLO v3",
  "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/64856",
  "author_name": "",
  "post_date": "2018-09-03T10:34:46.374522300Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I just wanted to know if there is anyway I could retrain YOLO weights regarding this competition. If there is one, please do let me know!</p>",
  "messages": [
    {
      "id": "380725",
      "postDate": "09/03/2018 10:34:46",
      "content": "<p>I just wanted to know if there is anyway I could retrain YOLO weights regarding this competition. If there is one, please do let me know!</p>",
      "rawMarkdown": "I just wanted to know if there is anyway I could retrain YOLO weights regarding this competition. If there is one, please do let me know!",
      "votes": null
    },
    {
      "id": "380752",
      "postDate": "09/03/2018 11:52:38",
      "content": "<p>Do you mean \"You Look Only One\" algorithm? In this case  you definitely can retrain YOLO algorithm corresponding this competition, and, I think, that it could be work better, then YOLO algorithm based only on pneumonia photo. I think, there is 2 way to do that:</p>\n\n<ol>\n<li>Take some marked data from object detection task and train your YOLO algorithm</li>\n<li>Take some marked data from just classification task from computer vision, then add some backgroung to image from dataset. As a result we will get pictures from background and picture from initial data in bounded boxes - based on this dataset we also can pretrain YOLO algorithm</li>\n</ol>",
      "rawMarkdown": "Do you mean \"You Look Only One\" algorithm? In this case  you definitely can retrain YOLO algorithm corresponding this competition, and, I think, that it could be work better, then YOLO algorithm based only on pneumonia photo. I think, there is 2 way to do that:\n\n 1. Take some marked data from object detection task and train your YOLO algorithm\n 2. Take some marked data from just classification task from computer vision, then add some backgroung to image from dataset. As a result we will get pictures from background and picture from initial data in bounded boxes - based on this dataset we also can pretrain YOLO algorithm",
      "votes": null
    },
    {
      "id": "380770",
      "postDate": "09/03/2018 12:25:18",
      "content": "<p>Option 1 looks easier, option 2 looks better. Let's see what i can do</p>",
      "rawMarkdown": "Option 1 looks easier, option 2 looks better. Let's see what i can do",
      "votes": null
    },
    {
      "id": "380892",
      "postDate": "09/03/2018 17:05:16",
      "content": "<p>I am yet to download the data for this competition but from some of my previous work I can say that you can also train YOLO from scratch : <a href=\"https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow\">https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow</a></p>",
      "rawMarkdown": "I am yet to download the data for this competition but from some of my previous work I can say that you can also train YOLO from scratch : https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow",
      "votes": null
    },
    {
      "id": "380928",
      "postDate": "09/03/2018 18:28:38",
      "content": "<p>Your Notebook served as an excellent example for the same! Thanks for that!</p>",
      "rawMarkdown": "Your Notebook served as an excellent example for the same! Thanks for that!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 380752,
      "author_name": "koza4ukdmitrij",
      "author_url": "",
      "post_date": "09/03/2018 11:52:38",
      "content": "<p>Do you mean \"You Look Only One\" algorithm? In this case  you definitely can retrain YOLO algorithm corresponding this competition, and, I think, that it could be work better, then YOLO algorithm based only on pneumonia photo. I think, there is 2 way to do that:</p>\n\n<ol>\n<li>Take some marked data from object detection task and train your YOLO algorithm</li>\n<li>Take some marked data from just classification task from computer vision, then add some backgroung to image from dataset. As a result we will get pictures from background and picture from initial data in bounded boxes - based on this dataset we also can pretrain YOLO algorithm</li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 380770,
          "author_name": "uds5501",
          "author_url": "",
          "post_date": "09/03/2018 12:25:18",
          "content": "<p>Option 1 looks easier, option 2 looks better. Let's see what i can do</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 380892,
      "author_name": "vijaybj",
      "author_url": "",
      "post_date": "09/03/2018 17:05:16",
      "content": "<p>I am yet to download the data for this competition but from some of my previous work I can say that you can also train YOLO from scratch : <a href=\"https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow\">https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 380928,
          "author_name": "uds5501",
          "author_url": "",
          "post_date": "09/03/2018 18:28:38",
          "content": "<p>Your Notebook served as an excellent example for the same! Thanks for that!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "380725": "I just wanted to know if there is anyway I could retrain YOLO weights regarding this competition. If there is one, please do let me know!",
    "380752": "Do you mean \"You Look Only One\" algorithm? In this case  you definitely can retrain YOLO algorithm corresponding this competition, and, I think, that it could be work better, then YOLO algorithm based only on pneumonia photo. I think, there is 2 way to do that:\n\n 1. Take some marked data from object detection task and train your YOLO algorithm\n 2. Take some marked data from just classification task from computer vision, then add some backgroung to image from dataset. As a result we will get pictures from background and picture from initial data in bounded boxes - based on this dataset we also can pretrain YOLO algorithm",
    "380770": "Option 1 looks easier, option 2 looks better. Let's see what i can do",
    "380892": "I am yet to download the data for this competition but from some of my previous work I can say that you can also train YOLO from scratch : https://www.kaggle.com/vijaybj/yolo-for-detection-of-bounding-boxes-tensorflow",
    "380928": "Your Notebook served as an excellent example for the same! Thanks for that!"
  },
  "source": "meta"
}