{
  "id": 190099,
  "title": "Some CNN network modeling suggestions",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/190099",
  "author_name": "Dewei Chen",
  "post_date": "2020-10-10T06:44:44.106000",
  "votes": 9,
  "comment_count": 1,
  "views": 0,
  "content": "<p>The top place popular CNN models:</p>\n<ol>\n<li><p>ResNet: is the basic CNN net structure, which can be used in any type of data and any competitions. such as: ResNet-50 , ResNet-101 or ResNet-152 is some example of ResNet. ResNet always have a well performance in the competition, and it could be one of component of your final stacking model. In<a href=\"https://www.kaggle.com/c/landmark-recognition-2019/overview\" target=\"_blank\"> 2019 Google Landmark recognition</a> and also <a href=\"https://www.kaggle.com/c/landmark-recognition-2020/discussion/187821\" target=\"_blank\">2020 Google Landmark recognition</a>  1st place solution used this network in final modeling.</p></li>\n<li><p>DenseNet: is deeper than ResNet and have a similar principle with ResNet. The key point of this network is the word \"Dense\".  DenseNet-121 is one of popular structure of it. And it get a well performance in <a href=\"https://www.kaggle.com/c/histopathologic-cancer-detection/overview\" target=\"_blank\">Cancer Detection Competition</a></p></li>\n<li><p>EfficientNet: This network is MOST popular in EVERY computer vision competitions. There are B0~B7, totally 8 main structure. That's one of the reason people loves Karas as well. Too much competitions using this model and get medal in the end. However, The top place solutions, always follows some other networks, because data scientist always have their specials thinking. <br>\nTherefore, EfficientNet is much suitable for middle place team like us.</p></li>\n</ol>\n<p>If there is any other popular network you think, please markdown, and I'll updating it.</p>",
  "messages": [
    {
      "id": 1044837,
      "postDate": "2020-10-10T06:44:44.107Z",
      "content": "<p>The top place popular CNN models:</p>\n<ol>\n<li><p>ResNet: is the basic CNN net structure, which can be used in any type of data and any competitions. such as: ResNet-50 , ResNet-101 or ResNet-152 is some example of ResNet. ResNet always have a well performance in the competition, and it could be one of component of your final stacking model. In<a href=\"https://www.kaggle.com/c/landmark-recognition-2019/overview\" target=\"_blank\"> 2019 Google Landmark recognition</a> and also <a href=\"https://www.kaggle.com/c/landmark-recognition-2020/discussion/187821\" target=\"_blank\">2020 Google Landmark recognition</a>  1st place solution used this network in final modeling.</p></li>\n<li><p>DenseNet: is deeper than ResNet and have a similar principle with ResNet. The key point of this network is the word \"Dense\".  DenseNet-121 is one of popular structure of it. And it get a well performance in <a href=\"https://www.kaggle.com/c/histopathologic-cancer-detection/overview\" target=\"_blank\">Cancer Detection Competition</a></p></li>\n<li><p>EfficientNet: This network is MOST popular in EVERY computer vision competitions. There are B0~B7, totally 8 main structure. That's one of the reason people loves Karas as well. Too much competitions using this model and get medal in the end. However, The top place solutions, always follows some other networks, because data scientist always have their specials thinking. <br>\nTherefore, EfficientNet is much suitable for middle place team like us.</p></li>\n</ol>\n<p>If there is any other popular network you think, please markdown, and I'll updating it.</p>",
      "rawMarkdown": "The top place popular CNN models:\n\n1. ResNet: is the basic CNN net structure, which can be used in any type of data and any competitions. such as: ResNet-50 , ResNet-101 or ResNet-152 is some example of ResNet. ResNet always have a well performance in the competition, and it could be one of component of your final stacking model. In[ 2019 Google Landmark recognition](https://www.kaggle.com/c/landmark-recognition-2019/overview) and also [2020 Google Landmark recognition](https://www.kaggle.com/c/landmark-recognition-2020/discussion/187821)  1st place solution used this network in final modeling.\n\n2. DenseNet: is deeper than ResNet and have a similar principle with ResNet. The key point of this network is the word \"Dense\".  DenseNet-121 is one of popular structure of it. And it get a well performance in [Cancer Detection Competition](https://www.kaggle.com/c/histopathologic-cancer-detection/overview)\n\n3. EfficientNet: This network is MOST popular in EVERY computer vision competitions. There are B0~B7, totally 8 main structure. That's one of the reason people loves Karas as well. Too much competitions using this model and get medal in the end. However, The top place solutions, always follows some other networks, because data scientist always have their specials thinking. \nTherefore, EfficientNet is much suitable for middle place team like us.\n\n\n\nIf there is any other popular network you think, please markdown, and I'll updating it.",
      "votes": 9
    },
    {
      "id": 1045373,
      "postDate": "2020-10-10T15:06:06.973Z",
      "content": "<p>Great work! Keep going.</p>",
      "rawMarkdown": "Great work! Keep going.",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1045373,
      "author_name": "Mobasshir Bhuiya Shagor",
      "author_url": "",
      "post_date": "2020-10-10T15:06:06.973000",
      "content": "<p>Great work! Keep going.</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1044837": "The top place popular CNN models:\n\n1. ResNet: is the basic CNN net structure, which can be used in any type of data and any competitions. such as: ResNet-50 , ResNet-101 or ResNet-152 is some example of ResNet. ResNet always have a well performance in the competition, and it could be one of component of your final stacking model. In[ 2019 Google Landmark recognition](https://www.kaggle.com/c/landmark-recognition-2019/overview) and also [2020 Google Landmark recognition](https://www.kaggle.com/c/landmark-recognition-2020/discussion/187821)  1st place solution used this network in final modeling.\n\n2. DenseNet: is deeper than ResNet and have a similar principle with ResNet. The key point of this network is the word \"Dense\".  DenseNet-121 is one of popular structure of it. And it get a well performance in [Cancer Detection Competition](https://www.kaggle.com/c/histopathologic-cancer-detection/overview)\n\n3. EfficientNet: This network is MOST popular in EVERY computer vision competitions. There are B0~B7, totally 8 main structure. That's one of the reason people loves Karas as well. Too much competitions using this model and get medal in the end. However, The top place solutions, always follows some other networks, because data scientist always have their specials thinking. \nTherefore, EfficientNet is much suitable for middle place team like us.\n\n\n\nIf there is any other popular network you think, please markdown, and I'll updating it.",
    "1045373": "Great work! Keep going."
  }
}