{
  "id": 290042,
  "title": "YOLOv5: Here we go again | LB: 0.45+",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/290042",
  "author_name": "",
  "post_date": "2021-11-22T23:25:19.806263Z",
  "votes": 29,
  "comment_count": 23,
  "views": 0,
  "content": "<p><img src=\"https://i.ibb.co/hgcdCJH/yolov5.jpg\" alt=\"yolov5\"></p>\n<p><strong>YOLOv5</strong> has become very popular and the is no doubt it will be a strong candidate here. Hence published notebooks for getting started. <strong>YOLOv5</strong> comes with <strong>W&amp;B</strong> integration by default hence you can track your training live and carry on many experiments without that much hassle … </p>\n<h2>Notebooks:</h2>\n<ul>\n<li>Train: <a href=\"https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-train\" target=\"_blank\">Great-Barrier-Reef: YOLOv5 [train] 🌊</a></li>\n<li>Infer: <a href=\"https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-infer\" target=\"_blank\">Great-Barrier-Reef: YOLOv5 [infer] 🌊</a></li>\n</ul>\n<blockquote>\n  <p>Inference here will not be like typical Object Detection competition as we have to use <strong>time-series API</strong></p>\n</blockquote>\n<h2>Training Samples:</h2>\n<p><img src=\"https://i.ibb.co/RQ0k1DZ/train-samples.png\" alt=\"train-samples\"></p>\n<h2>Validation Samples:</h2>\n<p><img src=\"https://i.ibb.co/my2fW37/valid-samples.png\" alt=\"valid-samples\"></p>\n<h2>History</h2>\n<p><img src=\"https://i.ibb.co/fxj1dMY/epoch-vs-score.png\" alt=\"epoch-vs-score\"></p>\n<h2>WandB</h2>\n<p><img src=\"https://i.ibb.co/rpFc8py/plot.png\" alt=\"plot\"></p>\n<p>Happy Kagglling :)</p>",
  "messages": [
    {
      "id": "1592138",
      "postDate": "11/22/2021 23:25:19",
      "content": "<p><img src=\"https://i.ibb.co/hgcdCJH/yolov5.jpg\" alt=\"yolov5\"></p>\n<p><strong>YOLOv5</strong> has become very popular and the is no doubt it will be a strong candidate here. Hence published notebooks for getting started. <strong>YOLOv5</strong> comes with <strong>W&amp;B</strong> integration by default hence you can track your training live and carry on many experiments without that much hassle … </p>\n<h2>Notebooks:</h2>\n<ul>\n<li>Train: <a href=\"https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-train\" target=\"_blank\">Great-Barrier-Reef: YOLOv5 [train] 🌊</a></li>\n<li>Infer: <a href=\"https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-infer\" target=\"_blank\">Great-Barrier-Reef: YOLOv5 [infer] 🌊</a></li>\n</ul>\n<blockquote>\n  <p>Inference here will not be like typical Object Detection competition as we have to use <strong>time-series API</strong></p>\n</blockquote>\n<h2>Training Samples:</h2>\n<p><img src=\"https://i.ibb.co/RQ0k1DZ/train-samples.png\" alt=\"train-samples\"></p>\n<h2>Validation Samples:</h2>\n<p><img src=\"https://i.ibb.co/my2fW37/valid-samples.png\" alt=\"valid-samples\"></p>\n<h2>History</h2>\n<p><img src=\"https://i.ibb.co/fxj1dMY/epoch-vs-score.png\" alt=\"epoch-vs-score\"></p>\n<h2>WandB</h2>\n<p><img src=\"https://i.ibb.co/rpFc8py/plot.png\" alt=\"plot\"></p>\n<p>Happy Kagglling :)</p>",
      "rawMarkdown": "<img src=\"https://i.ibb.co/hgcdCJH/yolov5.jpg\" alt=\"yolov5\" border=\"0\">\n\n**YOLOv5** has become very popular and the is no doubt it will be a strong candidate here. Hence published notebooks for getting started. **YOLOv5** comes with **W&B** integration by default hence you can track your training live and carry on many experiments without that much hassle ... \n\n## Notebooks:\n* Train: [Great-Barrier-Reef: YOLOv5 [train] 🌊](https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-train)\n* Infer: [Great-Barrier-Reef: YOLOv5 [infer] 🌊](https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-infer)\n\n> Inference here will not be like typical Object Detection competition as we have to use **time-series API**\n\n## Training Samples:\n<img src=\"https://i.ibb.co/RQ0k1DZ/train-samples.png\" alt=\"train-samples\" border=\"0\" width=500>\n\n## Validation Samples:\n<img src=\"https://i.ibb.co/my2fW37/valid-samples.png\" alt=\"valid-samples\" border=\"0\" width=500>\n\n## History\n<img src=\"https://i.ibb.co/fxj1dMY/epoch-vs-score.png\" alt=\"epoch-vs-score\" border=\"0\" width=500>\n\n## WandB\n<img src=\"https://i.ibb.co/rpFc8py/plot.png\" alt=\"plot\" border=\"0\">\n\nHappy Kagglling :)",
      "votes": null
    },
    {
      "id": "1592321",
      "postDate": "11/23/2021 03:28:31",
      "content": "<p>Hi, thanks for sharing!</p>",
      "rawMarkdown": "Hi, thanks for sharing!",
      "votes": null
    },
    {
      "id": "1594450",
      "postDate": "11/24/2021 20:59:23",
      "content": "<p>Thanks a lot!!</p>",
      "rawMarkdown": "Thanks a lot!!",
      "votes": null
    },
    {
      "id": "1594603",
      "postDate": "11/25/2021 02:28:27",
      "content": "<p>Thanks for sharing this</p>",
      "rawMarkdown": "Thanks for sharing this",
      "votes": null
    },
    {
      "id": "1594818",
      "postDate": "11/25/2021 06:24:49",
      "content": "<h2>Update: 25-11-2021</h2>\n<ul>\n<li>Had some minor bugs, fixed it</li>\n<li>LB: <code>0.453</code> which is currently the best public-notebook </li>\n</ul>",
      "rawMarkdown": "## Update: 25-11-2021\n* Had some minor bugs, fixed it\n* LB: `0.453` which is currently the best public-notebook",
      "votes": null
    },
    {
      "id": "1594981",
      "postDate": "11/25/2021 09:41:22",
      "content": "<p>Is Yolov5 allowed in this competition though? There were issues with it in past competitions</p>",
      "rawMarkdown": "Is Yolov5 allowed in this competition though? There were issues with it in past competitions",
      "votes": null
    },
    {
      "id": "1594993",
      "postDate": "11/25/2021 09:52:56",
      "content": "<p>I think you are talking about the <strong>GlobalWheatDetection</strong> Competition. There winning solution needed to be <strong>MIT</strong> but here, </p>\n<blockquote>\n  <p>WINNER LICENSE TYPE: Non-Exclusive</p>\n</blockquote>",
      "rawMarkdown": "I think you are talking about the **GlobalWheatDetection** Competition. There winning solution needed to be **MIT** but here, \n> WINNER LICENSE TYPE: Non-Exclusive",
      "votes": null
    },
    {
      "id": "1595506",
      "postDate": "11/25/2021 18:39:06",
      "content": "<p>Thanks for sharing <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>, do you think SSD based algorithm will also work here apart from YOLO?</p>",
      "rawMarkdown": "Thanks for sharing @awsaf49, do you think SSD based algorithm will also work here apart from YOLO?",
      "votes": null
    },
    {
      "id": "1595543",
      "postDate": "11/25/2021 18:58:36",
      "content": "<p>haven't tried them yet. Need to carry out more experiments to figure things out …</p>",
      "rawMarkdown": "haven't tried them yet. Need to carry out more experiments to figure things out ...",
      "votes": null
    },
    {
      "id": "1598981",
      "postDate": "11/29/2021 03:48:46",
      "content": "<p>Thanks afor sharing, it is a great work. <br>\nIf you don't mind, would you say which version of YOLOv5 [train] you used to get 0.453 LB？</p>",
      "rawMarkdown": "Thanks afor sharing, it is a great work. \nIf you don't mind, would you say which version of YOLOv5 [train] you used to get 0.453 LB？",
      "votes": null
    },
    {
      "id": "1599155",
      "postDate": "11/29/2021 07:22:39",
      "content": "<p>Version 13</p>",
      "rawMarkdown": "Version 13",
      "votes": null
    },
    {
      "id": "1599183",
      "postDate": "11/29/2021 07:57:24",
      "content": "<p>Got it, thanks.</p>",
      "rawMarkdown": "Got it, thanks.",
      "votes": null
    },
    {
      "id": "1599475",
      "postDate": "11/29/2021 13:14:35",
      "content": "<p>Thank you for the quick and rich introduction , it is really helpful 👍</p>",
      "rawMarkdown": "Thank you for the quick and rich introduction , it is really helpful 👍",
      "votes": null
    },
    {
      "id": "1600463",
      "postDate": "11/30/2021 12:49:03",
      "content": "<p>Your notebook is great. Thank you for inspiring me to create YoloX full training and inference pipeline for COTS dataset. Any feedback welcomed: <a href=\"https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset\" target=\"_blank\">https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset</a></p>",
      "rawMarkdown": "Your notebook is great. Thank you for inspiring me to create YoloX full training and inference pipeline for COTS dataset. Any feedback welcomed: https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset",
      "votes": null
    },
    {
      "id": "1600468",
      "postDate": "11/30/2021 12:52:47",
      "content": "<p>Superb Work <a href=\"https://www.kaggle.com/remekkinas\" target=\"_blank\">@remekkinas</a>. I was planning to write one for <strong>YOLOX</strong> I guess now I can reuse your notebook .. :)</p>",
      "rawMarkdown": "Superb Work @remekkinas. I was planning to write one for **YOLOX** I guess now I can reuse your notebook .. :)",
      "votes": null
    },
    {
      "id": "1600475",
      "postDate": "11/30/2021 12:57:30",
      "content": "<p>Sure. It works but still looking why loss is unstable …. Plase use this notebook and maybe you can find where the problem is.</p>",
      "rawMarkdown": "Sure. It works but still looking why loss is unstable .... Plase use this notebook and maybe you can find where the problem is.",
      "votes": null
    },
    {
      "id": "1600495",
      "postDate": "11/30/2021 13:17:41",
      "content": "<p>I see …. needs more epoch … as I can see YOLOX guys said that … more epoch is needed …. Let's try :)</p>",
      "rawMarkdown": "I see .... needs more epoch ... as I can see YOLOX guys said that ... more epoch is needed .... Let's try :)",
      "votes": null
    },
    {
      "id": "1600644",
      "postDate": "11/30/2021 15:27:16",
      "content": "<p>more epoch is usually always the answer 😃</p>",
      "rawMarkdown": "more epoch is usually always the answer 😃",
      "votes": null
    },
    {
      "id": "1600648",
      "postDate": "11/30/2021 15:31:46",
      "content": "<p>Yes. 😄 I am wondering why lost is such unstable and … trying to find answer. YOLOX creators said that longer training time is needed so I gave it time and now training for 100 epochs. I looked into my coco json converter and improved it a little bit (but I have not found any issues). Then I will try to set up training hyperparameters.  </p>",
      "rawMarkdown": "Yes. 😄 I am wondering why lost is such unstable and ... trying to find answer. YOLOX creators said that longer training time is needed so I gave it time and now training for 100 epochs. I looked into my coco json converter and improved it a little bit (but I have not found any issues). Then I will try to set up training hyperparameters.",
      "votes": null
    },
    {
      "id": "1603198",
      "postDate": "12/02/2021 11:49:54",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>  … I created inference part as well: <a href=\"https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots\" target=\"_blank\">https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots</a> </p>\n<p>If you have any feedback let me know … I will be more then happy. Now it is time to build strong model :)</p>",
      "rawMarkdown": "Hi @awsaf49  ... I created inference part as well: https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots \n\nIf you have any feedback let me know ... I will be more then happy. Now it is time to build strong model :)",
      "votes": null
    },
    {
      "id": "1638920",
      "postDate": "01/05/2022 06:56:40",
      "content": "<p><a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>  are you able to reach more than 0.5 with yolov5.  Not able to figure out why it is not able to match with correspondin YoloX. YoloX is very less featured compared YoloV5</p>",
      "rawMarkdown": "awsaf49  are you able to reach more than 0.5 with yolov5.  Not able to figure out why it is not able to match with correspondin YoloX. YoloX is very less featured compared YoloV5",
      "votes": null
    },
    {
      "id": "1639124",
      "postDate": "01/05/2022 11:58:50",
      "content": "<p>Not yet <a href=\"https://www.kaggle.com/jaideepvalani\" target=\"_blank\">@jaideepvalani</a> </p>",
      "rawMarkdown": "Not yet @jaideepvalani",
      "votes": null
    },
    {
      "id": "1657145",
      "postDate": "01/20/2022 00:21:28",
      "content": "<p>Thank for sharing the notebook!!! I love your kernels!! </p>",
      "rawMarkdown": "Thank for sharing the notebook!!! I love your kernels!!",
      "votes": null
    },
    {
      "id": "1658825",
      "postDate": "01/21/2022 10:37:55",
      "content": "<p>Thank you for clarifying my question! <a href=\"https://www.kaggle.com/Awsaf\" target=\"_blank\">@Awsaf</a>!</p>",
      "rawMarkdown": "Thank you for clarifying my question! @Awsaf!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1592321,
      "author_name": "kakarroto",
      "author_url": "",
      "post_date": "11/23/2021 03:28:31",
      "content": "<p>Hi, thanks for sharing!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1594450,
      "author_name": "ronlynes",
      "author_url": "",
      "post_date": "11/24/2021 20:59:23",
      "content": "<p>Thanks a lot!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1594603,
      "author_name": "dhinaharp",
      "author_url": "",
      "post_date": "11/25/2021 02:28:27",
      "content": "<p>Thanks for sharing this</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1594818,
      "author_name": "awsaf49",
      "author_url": "",
      "post_date": "11/25/2021 06:24:49",
      "content": "<h2>Update: 25-11-2021</h2>\n<ul>\n<li>Had some minor bugs, fixed it</li>\n<li>LB: <code>0.453</code> which is currently the best public-notebook </li>\n</ul>",
      "votes": null,
      "replies": [
        {
          "id": 1595506,
          "author_name": "saurabhbagchi",
          "author_url": "",
          "post_date": "11/25/2021 18:39:06",
          "content": "<p>Thanks for sharing <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>, do you think SSD based algorithm will also work here apart from YOLO?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1595543,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "11/25/2021 18:58:36",
          "content": "<p>haven't tried them yet. Need to carry out more experiments to figure things out …</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1638920,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "01/05/2022 06:56:40",
          "content": "<p><a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>  are you able to reach more than 0.5 with yolov5.  Not able to figure out why it is not able to match with correspondin YoloX. YoloX is very less featured compared YoloV5</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1639124,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "01/05/2022 11:58:50",
          "content": "<p>Not yet <a href=\"https://www.kaggle.com/jaideepvalani\" target=\"_blank\">@jaideepvalani</a> </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1594981,
      "author_name": "matthewmasters",
      "author_url": "",
      "post_date": "11/25/2021 09:41:22",
      "content": "<p>Is Yolov5 allowed in this competition though? There were issues with it in past competitions</p>",
      "votes": null,
      "replies": [
        {
          "id": 1594993,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "11/25/2021 09:52:56",
          "content": "<p>I think you are talking about the <strong>GlobalWheatDetection</strong> Competition. There winning solution needed to be <strong>MIT</strong> but here, </p>\n<blockquote>\n  <p>WINNER LICENSE TYPE: Non-Exclusive</p>\n</blockquote>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1658825,
          "author_name": "fuckvenkatraman",
          "author_url": "",
          "post_date": "01/21/2022 10:37:55",
          "content": "<p>Thank you for clarifying my question! <a href=\"https://www.kaggle.com/Awsaf\" target=\"_blank\">@Awsaf</a>!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1598981,
      "author_name": "garybios",
      "author_url": "",
      "post_date": "11/29/2021 03:48:46",
      "content": "<p>Thanks afor sharing, it is a great work. <br>\nIf you don't mind, would you say which version of YOLOv5 [train] you used to get 0.453 LB？</p>",
      "votes": null,
      "replies": [
        {
          "id": 1599155,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "11/29/2021 07:22:39",
          "content": "<p>Version 13</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1599183,
          "author_name": "garybios",
          "author_url": "",
          "post_date": "11/29/2021 07:57:24",
          "content": "<p>Got it, thanks.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1599475,
      "author_name": "mhslearner",
      "author_url": "",
      "post_date": "11/29/2021 13:14:35",
      "content": "<p>Thank you for the quick and rich introduction , it is really helpful 👍</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1600463,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "11/30/2021 12:49:03",
      "content": "<p>Your notebook is great. Thank you for inspiring me to create YoloX full training and inference pipeline for COTS dataset. Any feedback welcomed: <a href=\"https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset\" target=\"_blank\">https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1600468,
          "author_name": "awsaf49",
          "author_url": "",
          "post_date": "11/30/2021 12:52:47",
          "content": "<p>Superb Work <a href=\"https://www.kaggle.com/remekkinas\" target=\"_blank\">@remekkinas</a>. I was planning to write one for <strong>YOLOX</strong> I guess now I can reuse your notebook .. :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1600475,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "11/30/2021 12:57:30",
          "content": "<p>Sure. It works but still looking why loss is unstable …. Plase use this notebook and maybe you can find where the problem is.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1600495,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "11/30/2021 13:17:41",
          "content": "<p>I see …. needs more epoch … as I can see YOLOX guys said that … more epoch is needed …. Let's try :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1600644,
          "author_name": "drhabib",
          "author_url": "",
          "post_date": "11/30/2021 15:27:16",
          "content": "<p>more epoch is usually always the answer 😃</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1600648,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "11/30/2021 15:31:46",
          "content": "<p>Yes. 😄 I am wondering why lost is such unstable and … trying to find answer. YOLOX creators said that longer training time is needed so I gave it time and now training for 100 epochs. I looked into my coco json converter and improved it a little bit (but I have not found any issues). Then I will try to set up training hyperparameters.  </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1603198,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "12/02/2021 11:49:54",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/awsaf49\" target=\"_blank\">@awsaf49</a>  … I created inference part as well: <a href=\"https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots\" target=\"_blank\">https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots</a> </p>\n<p>If you have any feedback let me know … I will be more then happy. Now it is time to build strong model :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1657145,
      "author_name": "eugeneryu",
      "author_url": "",
      "post_date": "01/20/2022 00:21:28",
      "content": "<p>Thank for sharing the notebook!!! I love your kernels!! </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1592138": "<img src=\"https://i.ibb.co/hgcdCJH/yolov5.jpg\" alt=\"yolov5\" border=\"0\">\n\n**YOLOv5** has become very popular and the is no doubt it will be a strong candidate here. Hence published notebooks for getting started. **YOLOv5** comes with **W&B** integration by default hence you can track your training live and carry on many experiments without that much hassle ... \n\n## Notebooks:\n* Train: [Great-Barrier-Reef: YOLOv5 [train] 🌊](https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-train)\n* Infer: [Great-Barrier-Reef: YOLOv5 [infer] 🌊](https://www.kaggle.com/awsaf49/great-barrier-reef-yolov5-infer)\n\n> Inference here will not be like typical Object Detection competition as we have to use **time-series API**\n\n## Training Samples:\n<img src=\"https://i.ibb.co/RQ0k1DZ/train-samples.png\" alt=\"train-samples\" border=\"0\" width=500>\n\n## Validation Samples:\n<img src=\"https://i.ibb.co/my2fW37/valid-samples.png\" alt=\"valid-samples\" border=\"0\" width=500>\n\n## History\n<img src=\"https://i.ibb.co/fxj1dMY/epoch-vs-score.png\" alt=\"epoch-vs-score\" border=\"0\" width=500>\n\n## WandB\n<img src=\"https://i.ibb.co/rpFc8py/plot.png\" alt=\"plot\" border=\"0\">\n\nHappy Kagglling :)",
    "1592321": "Hi, thanks for sharing!",
    "1594450": "Thanks a lot!!",
    "1594603": "Thanks for sharing this",
    "1594818": "## Update: 25-11-2021\n* Had some minor bugs, fixed it\n* LB: `0.453` which is currently the best public-notebook",
    "1594981": "Is Yolov5 allowed in this competition though? There were issues with it in past competitions",
    "1594993": "I think you are talking about the **GlobalWheatDetection** Competition. There winning solution needed to be **MIT** but here, \n> WINNER LICENSE TYPE: Non-Exclusive",
    "1595506": "Thanks for sharing @awsaf49, do you think SSD based algorithm will also work here apart from YOLO?",
    "1595543": "haven't tried them yet. Need to carry out more experiments to figure things out ...",
    "1598981": "Thanks afor sharing, it is a great work. \nIf you don't mind, would you say which version of YOLOv5 [train] you used to get 0.453 LB？",
    "1599155": "Version 13",
    "1599183": "Got it, thanks.",
    "1599475": "Thank you for the quick and rich introduction , it is really helpful 👍",
    "1600463": "Your notebook is great. Thank you for inspiring me to create YoloX full training and inference pipeline for COTS dataset. Any feedback welcomed: https://www.kaggle.com/remekkinas/yolox-full-training-pipeline-for-cots-dataset",
    "1600468": "Superb Work @remekkinas. I was planning to write one for **YOLOX** I guess now I can reuse your notebook .. :)",
    "1600475": "Sure. It works but still looking why loss is unstable .... Plase use this notebook and maybe you can find where the problem is.",
    "1600495": "I see .... needs more epoch ... as I can see YOLOX guys said that ... more epoch is needed .... Let's try :)",
    "1600644": "more epoch is usually always the answer 😃",
    "1600648": "Yes. 😄 I am wondering why lost is such unstable and ... trying to find answer. YOLOX creators said that longer training time is needed so I gave it time and now training for 100 epochs. I looked into my coco json converter and improved it a little bit (but I have not found any issues). Then I will try to set up training hyperparameters.",
    "1603198": "Hi @awsaf49  ... I created inference part as well: https://www.kaggle.com/remekkinas/yolox-inference-on-kaggle-for-cots \n\nIf you have any feedback let me know ... I will be more then happy. Now it is time to build strong model :)",
    "1638920": "awsaf49  are you able to reach more than 0.5 with yolov5.  Not able to figure out why it is not able to match with correspondin YoloX. YoloX is very less featured compared YoloV5",
    "1639124": "Not yet @jaideepvalani",
    "1657145": "Thank for sharing the notebook!!! I love your kernels!!",
    "1658825": "Thank you for clarifying my question! @Awsaf!"
  },
  "source": "meta"
}