{
  "id": 500867,
  "title": "bird-vocalization-classifier",
  "url": "/competitions/birdclef-2024/discussion/500867",
  "author_name": "Tibor Vansa",
  "post_date": "2024-05-07T08:11:17.440000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I found this notebook with using google trained bird model<br>\n<a href=\"https://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook\" target=\"_blank\">https://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook</a><br>\nbut it is too slow. The obvious solution learned from previous year will be to convert it into onnx or openvino to run faster, but I failed to do it by <br>\ntf2onnx.convert.from_keras(model, input_signature)<br>\nHow to do it?  And if you succeded, did it help to get into time limit?<br>\nI did not found any details about model architecture, except that it should be EfficientNet B1. So why it is than so slow, when the pined starter solution is using efficientnetv2_b2, which is slightly bigger model?<br>\nThx</p>",
  "messages": [
    {
      "id": 2798409,
      "postDate": "2024-05-07T08:11:17.440Z",
      "content": "<p>I found this notebook with using google trained bird model<br>\n<a href=\"https://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook\" target=\"_blank\">https://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook</a><br>\nbut it is too slow. The obvious solution learned from previous year will be to convert it into onnx or openvino to run faster, but I failed to do it by <br>\ntf2onnx.convert.from_keras(model, input_signature)<br>\nHow to do it?  And if you succeded, did it help to get into time limit?<br>\nI did not found any details about model architecture, except that it should be EfficientNet B1. So why it is than so slow, when the pined starter solution is using efficientnetv2_b2, which is slightly bigger model?<br>\nThx</p>",
      "rawMarkdown": "I found this notebook with using google trained bird model\nhttps://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook\nbut it is too slow. The obvious solution learned from previous year will be to convert it into onnx or openvino to run faster, but I failed to do it by \ntf2onnx.convert.from_keras(model, input_signature)\nHow to do it?  And if you succeded, did it help to get into time limit?\nI did not found any details about model architecture, except that it should be EfficientNet B1. So why it is than so slow, when the pined starter solution is using efficientnetv2_b2, which is slightly bigger model?\nThx",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2798409": "I found this notebook with using google trained bird model\nhttps://www.kaggle.com/code/tabassumnova/google-bird-vocalization-classifier-model-77-acc/notebook\nbut it is too slow. The obvious solution learned from previous year will be to convert it into onnx or openvino to run faster, but I failed to do it by \ntf2onnx.convert.from_keras(model, input_signature)\nHow to do it?  And if you succeded, did it help to get into time limit?\nI did not found any details about model architecture, except that it should be EfficientNet B1. So why it is than so slow, when the pined starter solution is using efficientnetv2_b2, which is slightly bigger model?\nThx"
  }
}