{
  "id": 47798,
  "title": "Easily get 96.98% accuracy using MXNet",
  "url": "/competitions/cifar-10/discussion/47798",
  "author_name": "Sinyer Atlantis",
  "post_date": "2018-01-19T01:44:34.214000",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>We have created a series of cnn models for cifar10 image classification. With mxnet's gluon interface, it's easy to use a single model to reach 95.96% classification accuracy on cifar10 datasets, with very little time spent. </p>\n\n<p><img src=\"https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/single.png\" alt=\"\"></p>\n\n<p>Following is the model test results for the cifar10 raw dataset: </p>\n\n<p><img src=\"https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/ensemble.png\" alt=\"\"></p>\n\n<p>GPU: GTX 1070</p>\n\n<p>The reason gluon is chosen is because gluon's simplicity and efficiency. The gluon-based implementation of the same model is always faster than the corresponding keras version and get higher accuracy, and the model building is also simpler. </p>\n\n<p>Gluon and mxnet installation method is also very simple, refer to mxnet gluon tutorial can be easily done. \nDetailed parameters and code are on my <a href=\"https://github.com/SinyerAtlantis/deep_learning_gluon/tree/master/2.%20cnn_cifar10\">github</a>. </p>\n\n<p>The main idea and detailed discussion is from <a href=\"https://discuss.gluon.ai/t/topic/1545/423\">https://discuss.gluon.ai/t/topic/1545/423</a>. </p>\n\n<p>Related paper：</p>\n\n<ol>\n<li><a href=\"http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf\">Gradient-Based Learning Applied to Document Recognition</a></li>\n<li><a href=\"https://arxiv.org/abs/1512.03385\">Deep Residual Learning for Image Recognition</a></li>\n<li><a href=\"https://arxiv.org/abs/1603.05027\">Identity Mappings in Deep Residual Networks</a></li>\n<li><a href=\"https://arxiv.org/abs/1605.07146\">Wide Residual Networks</a></li>\n<li><a href=\"https://arxiv.org/abs/1502.03167\">Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift</a></li>\n</ol>",
  "messages": [
    {
      "id": 1048730,
      "postDate": "2020-10-13T18:30:04.493Z",
      "content": "<p>Thanks a lot for sharing this post, but the Github link you provided is invalid. Can you please fix it?</p>",
      "rawMarkdown": "Thanks a lot for sharing this post, but the Github link you provided is invalid. Can you please fix it?",
      "votes": 1,
      "replies": [
        {
          "id": 1069872,
          "postDate": "2020-11-05T03:04:36.647Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 270809,
      "postDate": "2018-01-19T01:44:34.213Z",
      "content": "<p>We have created a series of cnn models for cifar10 image classification. With mxnet's gluon interface, it's easy to use a single model to reach 95.96% classification accuracy on cifar10 datasets, with very little time spent. </p>\n\n<p><img src=\"https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/single.png\" alt=\"\"></p>\n\n<p>Following is the model test results for the cifar10 raw dataset: </p>\n\n<p><img src=\"https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/ensemble.png\" alt=\"\"></p>\n\n<p>GPU: GTX 1070</p>\n\n<p>The reason gluon is chosen is because gluon's simplicity and efficiency. The gluon-based implementation of the same model is always faster than the corresponding keras version and get higher accuracy, and the model building is also simpler. </p>\n\n<p>Gluon and mxnet installation method is also very simple, refer to mxnet gluon tutorial can be easily done. \nDetailed parameters and code are on my <a href=\"https://github.com/SinyerAtlantis/deep_learning_gluon/tree/master/2.%20cnn_cifar10\">github</a>. </p>\n\n<p>The main idea and detailed discussion is from <a href=\"https://discuss.gluon.ai/t/topic/1545/423\">https://discuss.gluon.ai/t/topic/1545/423</a>. </p>\n\n<p>Related paper：</p>\n\n<ol>\n<li><a href=\"http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf\">Gradient-Based Learning Applied to Document Recognition</a></li>\n<li><a href=\"https://arxiv.org/abs/1512.03385\">Deep Residual Learning for Image Recognition</a></li>\n<li><a href=\"https://arxiv.org/abs/1603.05027\">Identity Mappings in Deep Residual Networks</a></li>\n<li><a href=\"https://arxiv.org/abs/1605.07146\">Wide Residual Networks</a></li>\n<li><a href=\"https://arxiv.org/abs/1502.03167\">Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift</a></li>\n</ol>",
      "rawMarkdown": "We have created a series of cnn models for cifar10 image classification. With mxnet's gluon interface, it's easy to use a single model to reach 95.96% classification accuracy on cifar10 datasets, with very little time spent. \n\n![][1]\n\nFollowing is the model test results for the cifar10 raw dataset: \n\n![][2]\n\nGPU: GTX 1070\n\nThe reason gluon is chosen is because gluon's simplicity and efficiency. The gluon-based implementation of the same model is always faster than the corresponding keras version and get higher accuracy, and the model building is also simpler. \n\nGluon and mxnet installation method is also very simple, refer to mxnet gluon tutorial can be easily done. \nDetailed parameters and code are on my [github](https://github.com/SinyerAtlantis/deep_learning_gluon/tree/master/2.%20cnn_cifar10). \n\nThe main idea and detailed discussion is from https://discuss.gluon.ai/t/topic/1545/423. \n\nRelated paper：\n\n1. [Gradient-Based Learning Applied to Document Recognition][8]\n2. [Deep Residual Learning for Image Recognition][3]\n3. [Identity Mappings in Deep Residual Networks][4]\n4. [Wide Residual Networks][5]\n5. [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift][6]\n\n  [1]: https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/single.png\n  [2]: https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/ensemble.png\n[3]: https://arxiv.org/abs/1512.03385\n[4]: https://arxiv.org/abs/1603.05027\n[5]: https://arxiv.org/abs/1605.07146\n[6]: https://arxiv.org/abs/1502.03167\n[8]: http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf",
      "votes": 7
    }
  ],
  "comments": [
    {
      "id": 1048730,
      "author_name": "Yegane",
      "author_url": "",
      "post_date": "2020-10-13T18:30:04.493000",
      "content": "<p>Thanks a lot for sharing this post, but the Github link you provided is invalid. Can you please fix it?</p>",
      "votes": 1,
      "replies": [
        {
          "id": 1069872,
          "author_name": "",
          "author_url": "",
          "post_date": "2020-11-05T03:04:36.647000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1048730": "Thanks a lot for sharing this post, but the Github link you provided is invalid. Can you please fix it?",
    "270809": "We have created a series of cnn models for cifar10 image classification. With mxnet's gluon interface, it's easy to use a single model to reach 95.96% classification accuracy on cifar10 datasets, with very little time spent. \n\n![][1]\n\nFollowing is the model test results for the cifar10 raw dataset: \n\n![][2]\n\nGPU: GTX 1070\n\nThe reason gluon is chosen is because gluon's simplicity and efficiency. The gluon-based implementation of the same model is always faster than the corresponding keras version and get higher accuracy, and the model building is also simpler. \n\nGluon and mxnet installation method is also very simple, refer to mxnet gluon tutorial can be easily done. \nDetailed parameters and code are on my [github](https://github.com/SinyerAtlantis/deep_learning_gluon/tree/master/2.%20cnn_cifar10). \n\nThe main idea and detailed discussion is from https://discuss.gluon.ai/t/topic/1545/423. \n\nRelated paper：\n\n1. [Gradient-Based Learning Applied to Document Recognition][8]\n2. [Deep Residual Learning for Image Recognition][3]\n3. [Identity Mappings in Deep Residual Networks][4]\n4. [Wide Residual Networks][5]\n5. [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift][6]\n\n  [1]: https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/single.png\n  [2]: https://raw.githubusercontent.com/SinyerAtlantis/deep_learning_gluon/master/2.%20cnn_cifar10/ensemble.png\n[3]: https://arxiv.org/abs/1512.03385\n[4]: https://arxiv.org/abs/1603.05027\n[5]: https://arxiv.org/abs/1605.07146\n[6]: https://arxiv.org/abs/1502.03167\n[8]: http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf"
  }
}