{
  "id": 301747,
  "title": "EfficientDet Pytorch Starter(Train & Infer)",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/301747",
  "author_name": "DeepUnderstanding",
  "post_date": "2022-01-19T07:27:42.435000",
  "votes": 16,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I was looking for SOTA OD models, and have seen efficientdet to be one of the best, and I didn't see any PyTorch notebook on this.<br>\nThe legendary notebook <a href=\"https://www.kaggle.com/shonenkov/training-efficientdet/notebook\" target=\"_blank\">https://www.kaggle.com/shonenkov/training-efficientdet/notebook</a> by <a href=\"https://www.kaggle.com/shonenkov\" target=\"_blank\">@shonenkov</a>, I found it was a little too long for me, and also a little outdated.<br>\nSo I made my own efficientdet starter pack!</p>\n<p>I have kept it as simple as possible, using the smallest backbones, even using a tiny image size(320 x 320) edit it as much as you can! <br>\n(Since I don't wanna use my Kaggle GPU, as I will be requiring it XD)</p>\n<p>I hope it helps.<br>\nLink:<a href=\"https://www.kaggle.com/mrinath/efficientdet-train-pytorch\" target=\"_blank\">https://www.kaggle.com/mrinath/efficientdet-train-pytorch</a></p>\n<p><strong>PS</strong>: The thing is that the infer notebook trained with the notebook is giving a score <strong>0.0</strong>, for obvious reasons, as its a very small backbone and uses a very small image size, so should I release it?Dont know how helpful it will be</p>\n<p><strong>UPDATE</strong>(21st jan) : infer note book : <a href=\"https://www.kaggle.com/mrinath/effdet-infer-pytorch\" target=\"_blank\">https://www.kaggle.com/mrinath/effdet-infer-pytorch</a></p>",
  "messages": [
    {
      "id": 1656220,
      "postDate": "2022-01-19T07:27:42.437Z",
      "content": "<p>I was looking for SOTA OD models, and have seen efficientdet to be one of the best, and I didn't see any PyTorch notebook on this.<br>\nThe legendary notebook <a href=\"https://www.kaggle.com/shonenkov/training-efficientdet/notebook\" target=\"_blank\">https://www.kaggle.com/shonenkov/training-efficientdet/notebook</a> by <a href=\"https://www.kaggle.com/shonenkov\" target=\"_blank\">@shonenkov</a>, I found it was a little too long for me, and also a little outdated.<br>\nSo I made my own efficientdet starter pack!</p>\n<p>I have kept it as simple as possible, using the smallest backbones, even using a tiny image size(320 x 320) edit it as much as you can! <br>\n(Since I don't wanna use my Kaggle GPU, as I will be requiring it XD)</p>\n<p>I hope it helps.<br>\nLink:<a href=\"https://www.kaggle.com/mrinath/efficientdet-train-pytorch\" target=\"_blank\">https://www.kaggle.com/mrinath/efficientdet-train-pytorch</a></p>\n<p><strong>PS</strong>: The thing is that the infer notebook trained with the notebook is giving a score <strong>0.0</strong>, for obvious reasons, as its a very small backbone and uses a very small image size, so should I release it?Dont know how helpful it will be</p>\n<p><strong>UPDATE</strong>(21st jan) : infer note book : <a href=\"https://www.kaggle.com/mrinath/effdet-infer-pytorch\" target=\"_blank\">https://www.kaggle.com/mrinath/effdet-infer-pytorch</a></p>",
      "rawMarkdown": "I was looking for SOTA OD models, and have seen efficientdet to be one of the best, and I didn't see any PyTorch notebook on this.\nThe legendary notebook https://www.kaggle.com/shonenkov/training-efficientdet/notebook by @shonenkov, I found it was a little too long for me, and also a little outdated.\nSo I made my own efficientdet starter pack!\n\nI have kept it as simple as possible, using the smallest backbones, even using a tiny image size(320 x 320) edit it as much as you can! \n(Since I don't wanna use my Kaggle GPU, as I will be requiring it XD)\n\nI hope it helps.\nLink:https://www.kaggle.com/mrinath/efficientdet-train-pytorch\n\n**PS**: The thing is that the infer notebook trained with the notebook is giving a score **0.0**, for obvious reasons, as its a very small backbone and uses a very small image size, so should I release it?Dont know how helpful it will be\n\n**UPDATE**(21st jan) : infer note book : https://www.kaggle.com/mrinath/effdet-infer-pytorch\n",
      "votes": 16
    },
    {
      "id": 1658081,
      "postDate": "2022-01-20T17:56:58.187Z",
      "content": "<p><a href=\"https://www.kaggle.com/mrinath\" target=\"_blank\">@mrinath</a> my next experiment is with eff-det only. I will try to use D6/7 and compare with Yolov5 m6/x6. Thanks for providing the notebook, it is helpful. I suggest to make the inference public….will be helpful to people like me who are new to eff-det</p>",
      "rawMarkdown": "@mrinath my next experiment is with eff-det only. I will try to use D6/7 and compare with Yolov5 m6/x6. Thanks for providing the notebook, it is helpful. I suggest to make the inference public....will be helpful to people like me who are new to eff-det",
      "votes": 1,
      "replies": [
        {
          "id": 1658959,
          "postDate": "2022-01-21T12:37:58.567Z",
          "content": "<p>Thanks, Sanchit, I have done some improvement and now the score increased a bit from 0.0, it's not good but may help others.<br>\nthanks for the reply!</p>",
          "rawMarkdown": "Thanks, Sanchit, I have done some improvement and now the score increased a bit from 0.0, it's not good but may help others.\nthanks for the reply!",
          "votes": 1
        }
      ]
    },
    {
      "id": 1775138,
      "postDate": "2022-05-02T18:21:47.320Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1658081,
      "author_name": "Sanchit Vijay",
      "author_url": "",
      "post_date": "2022-01-20T17:56:58.187000",
      "content": "<p><a href=\"https://www.kaggle.com/mrinath\" target=\"_blank\">@mrinath</a> my next experiment is with eff-det only. I will try to use D6/7 and compare with Yolov5 m6/x6. Thanks for providing the notebook, it is helpful. I suggest to make the inference public….will be helpful to people like me who are new to eff-det</p>",
      "votes": 1,
      "replies": [
        {
          "id": 1658959,
          "author_name": "DeepUnderstanding",
          "author_url": "",
          "post_date": "2022-01-21T12:37:58.567000",
          "content": "<p>Thanks, Sanchit, I have done some improvement and now the score increased a bit from 0.0, it's not good but may help others.<br>\nthanks for the reply!</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 1775138,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-05-02T18:21:47.320000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1656220": "I was looking for SOTA OD models, and have seen efficientdet to be one of the best, and I didn't see any PyTorch notebook on this.\nThe legendary notebook https://www.kaggle.com/shonenkov/training-efficientdet/notebook by @shonenkov, I found it was a little too long for me, and also a little outdated.\nSo I made my own efficientdet starter pack!\n\nI have kept it as simple as possible, using the smallest backbones, even using a tiny image size(320 x 320) edit it as much as you can! \n(Since I don't wanna use my Kaggle GPU, as I will be requiring it XD)\n\nI hope it helps.\nLink:https://www.kaggle.com/mrinath/efficientdet-train-pytorch\n\n**PS**: The thing is that the infer notebook trained with the notebook is giving a score **0.0**, for obvious reasons, as its a very small backbone and uses a very small image size, so should I release it?Dont know how helpful it will be\n\n**UPDATE**(21st jan) : infer note book : https://www.kaggle.com/mrinath/effdet-infer-pytorch\n",
    "1658081": "@mrinath my next experiment is with eff-det only. I will try to use D6/7 and compare with Yolov5 m6/x6. Thanks for providing the notebook, it is helpful. I suggest to make the inference public....will be helpful to people like me who are new to eff-det",
    "1775138": ""
  }
}