{
  "id": 263789,
  "title": "Let me introduce EfficientnetV2 repository for Keras and TPU",
  "url": "/competitions/g2net-gravitational-wave-detection/discussion/263789",
  "author_name": "",
  "post_date": "2021-08-10T08:21:08.933430400Z",
  "votes": 15,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Many of the competitors use EfficientNet B7 in this competition.<br>\nRecently EfficientNet developers announced a network called EfficientNetV2.<br>\nThe paper is shown below.<br>\n<a href=\"https://arxiv.org/abs/2104.00298\" target=\"_blank\">EfficientNetV2: Smaller Models and Faster Training</a><br>\nAnd I found an excellent repository.<br>\n<a href=\"https://github.com/leondgarse/Keras_efficientnet_v2\" target=\"_blank\">Keras_efficientnet_v2</a><br>\nI trained the EfNetV2B0 with <a href=\"https://www.kaggle.com/miklgr500/cqt-g2net-efficientnetb7-tpu-training-w-b?scriptVersionId=67485043\" target=\"_blank\">the excellent baseline by Welf Crozzo</a>. The score was .862.<br>\nThis score is not good, but it may be possible to produce a better score by using a larger model (e.g. V2Large or V2XLarge) or by adding regularization.<br>\nTry to use the new network that may replace EfficientNetV1!<br>\nTips: V2 tends to be overfit.</p>",
  "messages": [
    {
      "id": "1463539",
      "postDate": "08/10/2021 08:21:08",
      "content": "<p>Many of the competitors use EfficientNet B7 in this competition.<br>\nRecently EfficientNet developers announced a network called EfficientNetV2.<br>\nThe paper is shown below.<br>\n<a href=\"https://arxiv.org/abs/2104.00298\" target=\"_blank\">EfficientNetV2: Smaller Models and Faster Training</a><br>\nAnd I found an excellent repository.<br>\n<a href=\"https://github.com/leondgarse/Keras_efficientnet_v2\" target=\"_blank\">Keras_efficientnet_v2</a><br>\nI trained the EfNetV2B0 with <a href=\"https://www.kaggle.com/miklgr500/cqt-g2net-efficientnetb7-tpu-training-w-b?scriptVersionId=67485043\" target=\"_blank\">the excellent baseline by Welf Crozzo</a>. The score was .862.<br>\nThis score is not good, but it may be possible to produce a better score by using a larger model (e.g. V2Large or V2XLarge) or by adding regularization.<br>\nTry to use the new network that may replace EfficientNetV1!<br>\nTips: V2 tends to be overfit.</p>",
      "rawMarkdown": "Many of the competitors use EfficientNet B7 in this competition.\nRecently EfficientNet developers announced a network called EfficientNetV2.\nThe paper is shown below.\n[EfficientNetV2: Smaller Models and Faster Training](https://arxiv.org/abs/2104.00298)\nAnd I found an excellent repository.\n[Keras_efficientnet_v2](https://github.com/leondgarse/Keras_efficientnet_v2)\nI trained the EfNetV2B0 with [the excellent baseline by Welf Crozzo](https://www.kaggle.com/miklgr500/cqt-g2net-efficientnetb7-tpu-training-w-b?scriptVersionId=67485043). The score was .862.\nThis score is not good, but it may be possible to produce a better score by using a larger model (e.g. V2Large or V2XLarge) or by adding regularization.\nTry to use the new network that may replace EfficientNetV1!\nTips: V2 tends to be overfit.",
      "votes": null
    },
    {
      "id": "1475925",
      "postDate": "08/16/2021 22:24:06",
      "content": "<p>I would also recommend the TF Hub for the TF/Keras users. </p>\n<p><a href=\"https://tfhub.dev/google/collections/efficientnet_v2/1\" target=\"_blank\">https://tfhub.dev/google/collections/efficientnet_v2/1</a></p>\n<p>It is pretty straightforward to define a new model based on the existing one.</p>",
      "rawMarkdown": "I would also recommend the TF Hub for the TF/Keras users. \n\nhttps://tfhub.dev/google/collections/efficientnet_v2/1\n\nIt is pretty straightforward to define a new model based on the existing one.",
      "votes": null
    },
    {
      "id": "1487654",
      "postDate": "08/23/2021 18:36:31",
      "content": "<p>I have used <a href=\"https://github.com/google/automl.git\" target=\"_blank\">this implementation</a> of EfficientNetV2 with TPU and it worked well for me in another competition.</p>",
      "rawMarkdown": "I have used [this implementation](https://github.com/google/automl.git) of EfficientNetV2 with TPU and it worked well for me in another competition.",
      "votes": null
    },
    {
      "id": "1488006",
      "postDate": "08/24/2021 02:43:48",
      "content": "<p>I know it, but I could not handle it well.<br>\nIt would help a lot of people if you published your baseline.</p>",
      "rawMarkdown": "I know it, but I could not handle it well.\nIt would help a lot of people if you published your baseline.",
      "votes": null
    },
    {
      "id": "1561187",
      "postDate": "10/27/2021 12:15:32",
      "content": "<p>Hey All,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "rawMarkdown": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1475925,
      "author_name": "snnclsr",
      "author_url": "",
      "post_date": "08/16/2021 22:24:06",
      "content": "<p>I would also recommend the TF Hub for the TF/Keras users. </p>\n<p><a href=\"https://tfhub.dev/google/collections/efficientnet_v2/1\" target=\"_blank\">https://tfhub.dev/google/collections/efficientnet_v2/1</a></p>\n<p>It is pretty straightforward to define a new model based on the existing one.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1487654,
      "author_name": "vzaguskin",
      "author_url": "",
      "post_date": "08/23/2021 18:36:31",
      "content": "<p>I have used <a href=\"https://github.com/google/automl.git\" target=\"_blank\">this implementation</a> of EfficientNetV2 with TPU and it worked well for me in another competition.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1488006,
          "author_name": "itsuki9180",
          "author_url": "",
          "post_date": "08/24/2021 02:43:48",
          "content": "<p>I know it, but I could not handle it well.<br>\nIt would help a lot of people if you published your baseline.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1561187,
      "author_name": "zerafachris",
      "author_url": "",
      "post_date": "10/27/2021 12:15:32",
      "content": "<p>Hey All,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1463539": "Many of the competitors use EfficientNet B7 in this competition.\nRecently EfficientNet developers announced a network called EfficientNetV2.\nThe paper is shown below.\n[EfficientNetV2: Smaller Models and Faster Training](https://arxiv.org/abs/2104.00298)\nAnd I found an excellent repository.\n[Keras_efficientnet_v2](https://github.com/leondgarse/Keras_efficientnet_v2)\nI trained the EfNetV2B0 with [the excellent baseline by Welf Crozzo](https://www.kaggle.com/miklgr500/cqt-g2net-efficientnetb7-tpu-training-w-b?scriptVersionId=67485043). The score was .862.\nThis score is not good, but it may be possible to produce a better score by using a larger model (e.g. V2Large or V2XLarge) or by adding regularization.\nTry to use the new network that may replace EfficientNetV1!\nTips: V2 tends to be overfit.",
    "1475925": "I would also recommend the TF Hub for the TF/Keras users. \n\nhttps://tfhub.dev/google/collections/efficientnet_v2/1\n\nIt is pretty straightforward to define a new model based on the existing one.",
    "1487654": "I have used [this implementation](https://github.com/google/automl.git) of EfficientNetV2 with TPU and it worked well for me in another competition.",
    "1488006": "I know it, but I could not handle it well.\nIt would help a lot of people if you published your baseline.",
    "1561187": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris"
  },
  "source": "meta"
}