{
  "id": 48470,
  "title": "improvements to resnet got 0.8933: single model, no data argument, no ensemble",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/48470",
  "author_name": "",
  "post_date": "2018-01-28T11:03:21.137387Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>my solutions for this competition: <a href=\"https://github.com/lifeiteng/TF_SpeechRecoChallenge\">https://github.com/lifeiteng/TF_SpeechRecoChallenge</a></p>",
  "messages": [
    {
      "id": "275117",
      "postDate": "01/28/2018 11:03:21",
      "content": "<p>my solutions for this competition: <a href=\"https://github.com/lifeiteng/TF_SpeechRecoChallenge\">https://github.com/lifeiteng/TF_SpeechRecoChallenge</a></p>",
      "rawMarkdown": "my solutions for this competition: https://github.com/lifeiteng/TF_SpeechRecoChallenge",
      "votes": null
    },
    {
      "id": "275595",
      "postDate": "01/29/2018 15:04:05",
      "content": "<p>Are you getting 0.8933 with a normal resnet? We had many models which can reach 0.89 without any problems like densenet, WRN-28-10D and resnext and so on. They can reach also 0.90 with mixup. See our repo: <a href=\"https://github.com/tugstugi/pytorch-speech-commands\">https://github.com/tugstugi/pytorch-speech-commands</a></p>\n\n<p>But we had always problems with normal resnets. Best score we could get was 87% public leaderboard.</p>",
      "rawMarkdown": "Are you getting 0.8933 with a normal resnet? We had many models which can reach 0.89 without any problems like densenet, WRN-28-10D and resnext and so on. They can reach also 0.90 with mixup. See our repo: https://github.com/tugstugi/pytorch-speech-commands\n\nBut we had always problems with normal resnets. Best score we could get was 87% public leaderboard.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 275595,
      "author_name": "tugstugi",
      "author_url": "",
      "post_date": "01/29/2018 15:04:05",
      "content": "<p>Are you getting 0.8933 with a normal resnet? We had many models which can reach 0.89 without any problems like densenet, WRN-28-10D and resnext and so on. They can reach also 0.90 with mixup. See our repo: <a href=\"https://github.com/tugstugi/pytorch-speech-commands\">https://github.com/tugstugi/pytorch-speech-commands</a></p>\n\n<p>But we had always problems with normal resnets. Best score we could get was 87% public leaderboard.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "275117": "my solutions for this competition: https://github.com/lifeiteng/TF_SpeechRecoChallenge",
    "275595": "Are you getting 0.8933 with a normal resnet? We had many models which can reach 0.89 without any problems like densenet, WRN-28-10D and resnext and so on. They can reach also 0.90 with mixup. See our repo: https://github.com/tugstugi/pytorch-speech-commands\n\nBut we had always problems with normal resnets. Best score we could get was 87% public leaderboard."
  },
  "source": "meta"
}