{
  "id": 74047,
  "title": "Might be a good starting point / code does not work",
  "url": "/competitions/inclusive-images-challenge/discussion/74047",
  "author_name": "Aadish Joshi",
  "post_date": "2018-12-07T22:13:46.109000",
  "votes": -3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>This is the code I tried which might be a good start to plan a strategy. The <strong>code does not work</strong> but I would be glad if someone tries on this lines further.</p>\n\n<p><strong>Strategy used:</strong>\nI found this in discussions: <a href=\"https://www.kaggle.com/victorhz/cnn-with-20-classes-trained-validation-set\">CNN with 20 Classes trained Validation Set</a> by Victor Zhao. So i decided to use that and fine tune it further.\nApproach used was to train 4 different models on the same training data and aggregating their prediction at the end. I used InceptionV3, InceptionResnetV2, Resnet50 and Victor zhao's model at the end, I ensemble the code to fine tune the accuracy. \nYou may find the code <a href=\"https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\">here</a>.\n<a href=\"https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\">https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code</a></p>",
  "messages": [
    {
      "id": 435348,
      "postDate": "2018-12-07T22:13:46.110Z",
      "content": "<p>This is the code I tried which might be a good start to plan a strategy. The <strong>code does not work</strong> but I would be glad if someone tries on this lines further.</p>\n\n<p><strong>Strategy used:</strong>\nI found this in discussions: <a href=\"https://www.kaggle.com/victorhz/cnn-with-20-classes-trained-validation-set\">CNN with 20 Classes trained Validation Set</a> by Victor Zhao. So i decided to use that and fine tune it further.\nApproach used was to train 4 different models on the same training data and aggregating their prediction at the end. I used InceptionV3, InceptionResnetV2, Resnet50 and Victor zhao's model at the end, I ensemble the code to fine tune the accuracy. \nYou may find the code <a href=\"https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\">here</a>.\n<a href=\"https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\">https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code</a></p>",
      "rawMarkdown": "This is the code I tried which might be a good start to plan a strategy. The **code does not work** but I would be glad if someone tries on this lines further.\n\n**Strategy used:**\nI found this in discussions: [CNN with 20 Classes trained Validation Set][1] by Victor Zhao. So i decided to use that and fine tune it further.\nApproach used was to train 4 different models on the same training data and aggregating their prediction at the end. I used InceptionV3, InceptionResnetV2, Resnet50 and Victor zhao's model at the end, I ensemble the code to fine tune the accuracy. \nYou may find the code [here][2].\nhttps://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\n\n\n  [1]: https://www.kaggle.com/victorhz/cnn-with-20-classes-trained-validation-set\n  [2]: https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code",
      "votes": -3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "435348": "This is the code I tried which might be a good start to plan a strategy. The **code does not work** but I would be glad if someone tries on this lines further.\n\n**Strategy used:**\nI found this in discussions: [CNN with 20 Classes trained Validation Set][1] by Victor Zhao. So i decided to use that and fine tune it further.\nApproach used was to train 4 different models on the same training data and aggregating their prediction at the end. I used InceptionV3, InceptionResnetV2, Resnet50 and Victor zhao's model at the end, I ensemble the code to fine tune the accuracy. \nYou may find the code [here][2].\nhttps://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code\n\n\n  [1]: https://www.kaggle.com/victorhz/cnn-with-20-classes-trained-validation-set\n  [2]: https://github.com/aadishjoshi/UniversalOpenImagePredictor/tree/master/00%20Code/Final%20Code"
  }
}