{
  "id": 335229,
  "title": "SOTA in segmentation example notebook",
  "url": "/competitions/hubmap-organ-segmentation/discussion/335229",
  "author_name": "",
  "post_date": "2022-07-05T08:30:56.787540Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p><strong>TL;DR</strong><br>\nCould you please help to create a a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. </p>\n<p>Your help would be highly appreciated!</p>\n<p><strong>In detail</strong></p>\n<p>To get good results in this competition, it makes sense to use the SOTA in segmentation. <br>\n<a href=\"https://paperswithcode.com/sota/instance-segmentation-on-coco-minival\" target=\"_blank\">Here</a> one can see that the SOTA nowadays is mask DINO, but in their GitHub page they just told me that they are going to upload their work only in a few months, so it'll probably happen after this competition is over.<br>\nHence, I decided to use the second-best segmentation model, which is SWIN-V2.</p>\n<p>In addition, to make sure I do not miss anything, I want to check the results on the original dataset (COCO) and make sure that the same results are obtained as reported in SWIN's paper.<br>\nI guess a full confirmation of this will take a long time, and Kaggle may stop and close in the middle, so I'm interested in doing it \"in parts\". That is, do only a prediction of segmentation of a single class / a few classes at a time, and in the end I will consolidate the results and make sure that it came out as I expected</p>\n<p>Hence, I am struggling for a couple of weeks now with creating a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. </p>\n<p>Could you please help to create one?</p>\n<p>PS: I have dozens of trials (based mainly on <a href=\"https://github.com/SwinTransformer/Swin-Transformer-Semantic-Segmentation\" target=\"_blank\">this</a>, <a href=\"https://colab.research.google.com/github/open-mmlab/mmsegmentation/blob/master/demo/MMSegmentation_Tutorial.ipynb\" target=\"_blank\">this</a>, <a href=\"https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/inference.md\" target=\"_blank\">this</a>, and <a href=\"https://www.kaggle.com/code/carnozhao/uwmgit-mmsegmentation-end-to-end-submission/notebook\" target=\"_blank\">this</a>), that I currently do not link here, as they all led to a dead end, so if you can supply me with a full working example Kaggle notebook, I will extremely appreciate that!</p>",
  "messages": [
    {
      "id": "1843970",
      "postDate": "07/05/2022 08:30:56",
      "content": "<p><strong>TL;DR</strong><br>\nCould you please help to create a a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. </p>\n<p>Your help would be highly appreciated!</p>\n<p><strong>In detail</strong></p>\n<p>To get good results in this competition, it makes sense to use the SOTA in segmentation. <br>\n<a href=\"https://paperswithcode.com/sota/instance-segmentation-on-coco-minival\" target=\"_blank\">Here</a> one can see that the SOTA nowadays is mask DINO, but in their GitHub page they just told me that they are going to upload their work only in a few months, so it'll probably happen after this competition is over.<br>\nHence, I decided to use the second-best segmentation model, which is SWIN-V2.</p>\n<p>In addition, to make sure I do not miss anything, I want to check the results on the original dataset (COCO) and make sure that the same results are obtained as reported in SWIN's paper.<br>\nI guess a full confirmation of this will take a long time, and Kaggle may stop and close in the middle, so I'm interested in doing it \"in parts\". That is, do only a prediction of segmentation of a single class / a few classes at a time, and in the end I will consolidate the results and make sure that it came out as I expected</p>\n<p>Hence, I am struggling for a couple of weeks now with creating a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. </p>\n<p>Could you please help to create one?</p>\n<p>PS: I have dozens of trials (based mainly on <a href=\"https://github.com/SwinTransformer/Swin-Transformer-Semantic-Segmentation\" target=\"_blank\">this</a>, <a href=\"https://colab.research.google.com/github/open-mmlab/mmsegmentation/blob/master/demo/MMSegmentation_Tutorial.ipynb\" target=\"_blank\">this</a>, <a href=\"https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/inference.md\" target=\"_blank\">this</a>, and <a href=\"https://www.kaggle.com/code/carnozhao/uwmgit-mmsegmentation-end-to-end-submission/notebook\" target=\"_blank\">this</a>), that I currently do not link here, as they all led to a dead end, so if you can supply me with a full working example Kaggle notebook, I will extremely appreciate that!</p>",
      "rawMarkdown": "**TL;DR**\nCould you please help to create a a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. \n\nYour help would be highly appreciated!\n\n**In detail**\n\nTo get good results in this competition, it makes sense to use the SOTA in segmentation. \n[Here](https://paperswithcode.com/sota/instance-segmentation-on-coco-minival) one can see that the SOTA nowadays is mask DINO, but in their GitHub page they just told me that they are going to upload their work only in a few months, so it'll probably happen after this competition is over.\nHence, I decided to use the second-best segmentation model, which is SWIN-V2.\n\nIn addition, to make sure I do not miss anything, I want to check the results on the original dataset (COCO) and make sure that the same results are obtained as reported in SWIN's paper.\nI guess a full confirmation of this will take a long time, and Kaggle may stop and close in the middle, so I'm interested in doing it \"in parts\". That is, do only a prediction of segmentation of a single class / a few classes at a time, and in the end I will consolidate the results and make sure that it came out as I expected\n\nHence, I am struggling for a couple of weeks now with creating a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. \n\nCould you please help to create one?\n\nPS: I have dozens of trials (based mainly on [this](https://github.com/SwinTransformer/Swin-Transformer-Semantic-Segmentation), [this](https://colab.research.google.com/github/open-mmlab/mmsegmentation/blob/master/demo/MMSegmentation_Tutorial.ipynb), [this](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/inference.md), and [this](https://www.kaggle.com/code/carnozhao/uwmgit-mmsegmentation-end-to-end-submission/notebook)), that I currently do not link here, as they all led to a dead end, so if you can supply me with a full working example Kaggle notebook, I will extremely appreciate that!",
      "votes": null
    },
    {
      "id": "1844963",
      "postDate": "07/06/2022 00:14:05",
      "content": "<p>What errors do you get? what seems to be your problems?</p>",
      "rawMarkdown": "What errors do you get? what seems to be your problems?",
      "votes": null
    },
    {
      "id": "1845205",
      "postDate": "07/06/2022 06:14:58",
      "content": "<p>Every time I change something, the exact error changes, but in general the errors teach me that my installations and\\or my setup have been done incorrectly…</p>",
      "rawMarkdown": "Every time I change something, the exact error changes, but in general the errors teach me that my installations and\\or my setup have been done incorrectly...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1844963,
      "author_name": "thedevastator",
      "author_url": "",
      "post_date": "07/06/2022 00:14:05",
      "content": "<p>What errors do you get? what seems to be your problems?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1845205,
          "author_name": "dudifrid",
          "author_url": "",
          "post_date": "07/06/2022 06:14:58",
          "content": "<p>Every time I change something, the exact error changes, but in general the errors teach me that my installations and\\or my setup have been done incorrectly…</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1843970": "**TL;DR**\nCould you please help to create a a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. \n\nYour help would be highly appreciated!\n\n**In detail**\n\nTo get good results in this competition, it makes sense to use the SOTA in segmentation. \n[Here](https://paperswithcode.com/sota/instance-segmentation-on-coco-minival) one can see that the SOTA nowadays is mask DINO, but in their GitHub page they just told me that they are going to upload their work only in a few months, so it'll probably happen after this competition is over.\nHence, I decided to use the second-best segmentation model, which is SWIN-V2.\n\nIn addition, to make sure I do not miss anything, I want to check the results on the original dataset (COCO) and make sure that the same results are obtained as reported in SWIN's paper.\nI guess a full confirmation of this will take a long time, and Kaggle may stop and close in the middle, so I'm interested in doing it \"in parts\". That is, do only a prediction of segmentation of a single class / a few classes at a time, and in the end I will consolidate the results and make sure that it came out as I expected\n\nHence, I am struggling for a couple of weeks now with creating a simple Kaggle notebook that predicts instance segmentation for COCO dataset for a single class using SWIN-V2, and finally shows the mean Average Precision on the whole class. \n\nCould you please help to create one?\n\nPS: I have dozens of trials (based mainly on [this](https://github.com/SwinTransformer/Swin-Transformer-Semantic-Segmentation), [this](https://colab.research.google.com/github/open-mmlab/mmsegmentation/blob/master/demo/MMSegmentation_Tutorial.ipynb), [this](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/inference.md), and [this](https://www.kaggle.com/code/carnozhao/uwmgit-mmsegmentation-end-to-end-submission/notebook)), that I currently do not link here, as they all led to a dead end, so if you can supply me with a full working example Kaggle notebook, I will extremely appreciate that!",
    "1844963": "What errors do you get? what seems to be your problems?",
    "1845205": "Every time I change something, the exact error changes, but in general the errors teach me that my installations and\\or my setup have been done incorrectly..."
  },
  "source": "meta"
}