{
  "id": 398776,
  "title": "Detect Annotations with EfficientDet",
  "url": "/competitions/benetech-making-graphs-accessible/discussion/398776",
  "author_name": "moth",
  "post_date": "2023-03-31T17:47:16.659000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've made two notebooks where I train and make predictions with an EfficientDet object detection network. It's implemented in PyTorch using Ross Wightman's <a href=\"https://github.com/rwightman/efficientdet-pytorch/\" target=\"_blank\">repository</a>.  You can find the notebooks here:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-train\" target=\"_blank\">Benetech | EfficientDet [Train]</a></li>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-inference\" target=\"_blank\">Benetech | EfficientDet [Inference]</a></li>\n</ul>\n<p>Using this network you can detect all the elements present in a chart. This is useful to:</p>\n<ul>\n<li>Detect the precise position of ticks and labels in both x-axis and y-axis.</li>\n<li>Once you know the position of the bounding boxes you can interpolate later the data series values.</li>\n<li>You can pass the bounding boxes to an OCR network to detect the text and numbers present in the image.</li>\n</ul>",
  "messages": [
    {
      "id": 2204600,
      "postDate": "2023-03-31T17:47:16.660Z",
      "content": "<p>I've made two notebooks where I train and make predictions with an EfficientDet object detection network. It's implemented in PyTorch using Ross Wightman's <a href=\"https://github.com/rwightman/efficientdet-pytorch/\" target=\"_blank\">repository</a>.  You can find the notebooks here:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-train\" target=\"_blank\">Benetech | EfficientDet [Train]</a></li>\n<li><a href=\"https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-inference\" target=\"_blank\">Benetech | EfficientDet [Inference]</a></li>\n</ul>\n<p>Using this network you can detect all the elements present in a chart. This is useful to:</p>\n<ul>\n<li>Detect the precise position of ticks and labels in both x-axis and y-axis.</li>\n<li>Once you know the position of the bounding boxes you can interpolate later the data series values.</li>\n<li>You can pass the bounding boxes to an OCR network to detect the text and numbers present in the image.</li>\n</ul>",
      "rawMarkdown": "I've made two notebooks where I train and make predictions with an EfficientDet object detection network. It's implemented in PyTorch using Ross Wightman's [repository](https://github.com/rwightman/efficientdet-pytorch/).  You can find the notebooks here:\n- [Benetech | EfficientDet [Train]](https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-train)\n- [Benetech | EfficientDet [Inference]](https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-inference)\n\nUsing this network you can detect all the elements present in a chart. This is useful to:\n- Detect the precise position of ticks and labels in both x-axis and y-axis.\n- Once you know the position of the bounding boxes you can interpolate later the data series values.\n- You can pass the bounding boxes to an OCR network to detect the text and numbers present in the image.\n\n",
      "votes": 6
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2204600": "I've made two notebooks where I train and make predictions with an EfficientDet object detection network. It's implemented in PyTorch using Ross Wightman's [repository](https://github.com/rwightman/efficientdet-pytorch/).  You can find the notebooks here:\n- [Benetech | EfficientDet [Train]](https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-train)\n- [Benetech | EfficientDet [Inference]](https://www.kaggle.com/code/alejopaullier/benetech-efficientdet-inference)\n\nUsing this network you can detect all the elements present in a chart. This is useful to:\n- Detect the precise position of ticks and labels in both x-axis and y-axis.\n- Once you know the position of the bounding boxes you can interpolate later the data series values.\n- You can pass the bounding boxes to an OCR network to detect the text and numbers present in the image.\n\n"
  }
}