{
  "id": 305367,
  "title": "Made a dataset of just the cropped COTS images 👍",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/305367",
  "author_name": "",
  "post_date": "2022-02-05T02:00:05.457651300Z",
  "votes": 24,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Hey all - </p>\n<p>I made a dataset of with all the cropped COTS images from the dataset based on the annotations in every video frame. </p>\n<ul>\n<li>Not immediately useful for the object detection models but could be used for data augmentation. * May also help some Kagglers practice building binary image classifiers before moving on to more complicated object detection models. </li>\n<li><strong>Dataset Link:</strong> <a href=\"https://www.kaggle.com/alexteboul/binary-cropped-crown-of-thorns-dataset\" target=\"_blank\">COTS v NotCOTS Cropped Crown of Thorns Dataset</a> (11,898 Cropped Tensorflow Kaggle Crown of Thorns Starfish from Bounding Boxes) </li>\n<li><strong>Code notebook to build the dataset:</strong> <a href=\"https://www.kaggle.com/alexteboul/cropped-crown-of-thorns-dataset-builder/\" target=\"_blank\">Cropped Crown of Thorns Dataset Builder</a></li>\n<li>Basically you have 11,898 images (all the cots from the videos) and 11,898 images of not COTS cropped from video frames that did not have annotations.</li>\n<li>There's also this User Guide Notebook for the new cropped images dataset: <a href=\"https://www.kaggle.com/alexteboul/user-guide-cots-v-notcots-dataset\" target=\"_blank\">User Guide COTS v NotCOTS Dataset</a></li>\n</ul>\n<p>Hope this helps!</p>\n<p>Joined the competition today without much time but let's see how this goes haha.</p>",
  "messages": [
    {
      "id": "1676463",
      "postDate": "02/05/2022 02:00:05",
      "content": "<p>Hey all - </p>\n<p>I made a dataset of with all the cropped COTS images from the dataset based on the annotations in every video frame. </p>\n<ul>\n<li>Not immediately useful for the object detection models but could be used for data augmentation. * May also help some Kagglers practice building binary image classifiers before moving on to more complicated object detection models. </li>\n<li><strong>Dataset Link:</strong> <a href=\"https://www.kaggle.com/alexteboul/binary-cropped-crown-of-thorns-dataset\" target=\"_blank\">COTS v NotCOTS Cropped Crown of Thorns Dataset</a> (11,898 Cropped Tensorflow Kaggle Crown of Thorns Starfish from Bounding Boxes) </li>\n<li><strong>Code notebook to build the dataset:</strong> <a href=\"https://www.kaggle.com/alexteboul/cropped-crown-of-thorns-dataset-builder/\" target=\"_blank\">Cropped Crown of Thorns Dataset Builder</a></li>\n<li>Basically you have 11,898 images (all the cots from the videos) and 11,898 images of not COTS cropped from video frames that did not have annotations.</li>\n<li>There's also this User Guide Notebook for the new cropped images dataset: <a href=\"https://www.kaggle.com/alexteboul/user-guide-cots-v-notcots-dataset\" target=\"_blank\">User Guide COTS v NotCOTS Dataset</a></li>\n</ul>\n<p>Hope this helps!</p>\n<p>Joined the competition today without much time but let's see how this goes haha.</p>",
      "rawMarkdown": "Hey all - \n\nI made a dataset of with all the cropped COTS images from the dataset based on the annotations in every video frame. \n* Not immediately useful for the object detection models but could be used for data augmentation. * May also help some Kagglers practice building binary image classifiers before moving on to more complicated object detection models. \n* **Dataset Link:** [COTS v NotCOTS Cropped Crown of Thorns Dataset](https://www.kaggle.com/alexteboul/binary-cropped-crown-of-thorns-dataset) (11,898 Cropped Tensorflow Kaggle Crown of Thorns Starfish from Bounding Boxes) \n* **Code notebook to build the dataset:** [Cropped Crown of Thorns Dataset Builder](https://www.kaggle.com/alexteboul/cropped-crown-of-thorns-dataset-builder/)\n* Basically you have 11,898 images (all the cots from the videos) and 11,898 images of not COTS cropped from video frames that did not have annotations.\n* There's also this User Guide Notebook for the new cropped images dataset: [User Guide COTS v NotCOTS Dataset](https://www.kaggle.com/alexteboul/user-guide-cots-v-notcots-dataset)\n\nHope this helps!\n\nJoined the competition today without much time but let's see how this goes haha.",
      "votes": null
    },
    {
      "id": "1676543",
      "postDate": "02/05/2022 04:39:51",
      "content": "<p>Thanks! I've been looking for something like this.</p>",
      "rawMarkdown": "Thanks! I've been looking for something like this.",
      "votes": null
    },
    {
      "id": "1676574",
      "postDate": "02/05/2022 05:15:07",
      "content": "<p>It can also be used to make some new data in negative image by copy-paste augment.</p>",
      "rawMarkdown": "It can also be used to make some new data in negative image by copy-paste augment.",
      "votes": null
    },
    {
      "id": "1677233",
      "postDate": "02/05/2022 15:38:24",
      "content": "<p>Sweet hope it helps!</p>",
      "rawMarkdown": "Sweet hope it helps!",
      "votes": null
    },
    {
      "id": "1677245",
      "postDate": "02/05/2022 15:42:49",
      "content": "<p>Yeah 👌 gonna make another dataset soon to match the format of the original competition dataset but with these crops randomly thrown all over those images + the .csv with annotation bounding boxes. We'll see if it helps or hurts LB though. Can't imagine any of these models are going to end up being that robust anyways.</p>",
      "rawMarkdown": "Yeah 👌 gonna make another dataset soon to match the format of the original competition dataset but with these crops randomly thrown all over those images + the .csv with annotation bounding boxes. We'll see if it helps or hurts LB though. Can't imagine any of these models are going to end up being that robust anyways.",
      "votes": null
    },
    {
      "id": "1677330",
      "postDate": "02/05/2022 16:37:51",
      "content": "<p>Thanks for making the dataset <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a>!</p>",
      "rawMarkdown": "Thanks for making the dataset @alexteboul!",
      "votes": null
    },
    {
      "id": "1678907",
      "postDate": "02/06/2022 21:00:17",
      "content": "<p>Thanks for your work <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a> </p>",
      "rawMarkdown": "Thanks for your work @alexteboul",
      "votes": null
    },
    {
      "id": "1679125",
      "postDate": "02/07/2022 02:20:46",
      "content": "<p>with enough cropped samples, then  detector + classifier may work well.</p>",
      "rawMarkdown": "with enough cropped samples, then  detector + classifier may work well.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1676543,
      "author_name": "kennyxie",
      "author_url": "",
      "post_date": "02/05/2022 04:39:51",
      "content": "<p>Thanks! I've been looking for something like this.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1677233,
          "author_name": "alexteboul",
          "author_url": "",
          "post_date": "02/05/2022 15:38:24",
          "content": "<p>Sweet hope it helps!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1676574,
      "author_name": "freshair1996",
      "author_url": "",
      "post_date": "02/05/2022 05:15:07",
      "content": "<p>It can also be used to make some new data in negative image by copy-paste augment.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1677245,
          "author_name": "alexteboul",
          "author_url": "",
          "post_date": "02/05/2022 15:42:49",
          "content": "<p>Yeah 👌 gonna make another dataset soon to match the format of the original competition dataset but with these crops randomly thrown all over those images + the .csv with annotation bounding boxes. We'll see if it helps or hurts LB though. Can't imagine any of these models are going to end up being that robust anyways.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1677330,
      "author_name": "bobliuuu",
      "author_url": "",
      "post_date": "02/05/2022 16:37:51",
      "content": "<p>Thanks for making the dataset <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a>!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1678907,
      "author_name": "nebipeker",
      "author_url": "",
      "post_date": "02/06/2022 21:00:17",
      "content": "<p>Thanks for your work <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1679125,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/07/2022 02:20:46",
      "content": "<p>with enough cropped samples, then  detector + classifier may work well.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1676463": "Hey all - \n\nI made a dataset of with all the cropped COTS images from the dataset based on the annotations in every video frame. \n* Not immediately useful for the object detection models but could be used for data augmentation. * May also help some Kagglers practice building binary image classifiers before moving on to more complicated object detection models. \n* **Dataset Link:** [COTS v NotCOTS Cropped Crown of Thorns Dataset](https://www.kaggle.com/alexteboul/binary-cropped-crown-of-thorns-dataset) (11,898 Cropped Tensorflow Kaggle Crown of Thorns Starfish from Bounding Boxes) \n* **Code notebook to build the dataset:** [Cropped Crown of Thorns Dataset Builder](https://www.kaggle.com/alexteboul/cropped-crown-of-thorns-dataset-builder/)\n* Basically you have 11,898 images (all the cots from the videos) and 11,898 images of not COTS cropped from video frames that did not have annotations.\n* There's also this User Guide Notebook for the new cropped images dataset: [User Guide COTS v NotCOTS Dataset](https://www.kaggle.com/alexteboul/user-guide-cots-v-notcots-dataset)\n\nHope this helps!\n\nJoined the competition today without much time but let's see how this goes haha.",
    "1676543": "Thanks! I've been looking for something like this.",
    "1676574": "It can also be used to make some new data in negative image by copy-paste augment.",
    "1677233": "Sweet hope it helps!",
    "1677245": "Yeah 👌 gonna make another dataset soon to match the format of the original competition dataset but with these crops randomly thrown all over those images + the .csv with annotation bounding boxes. We'll see if it helps or hurts LB though. Can't imagine any of these models are going to end up being that robust anyways.",
    "1677330": "Thanks for making the dataset @alexteboul!",
    "1678907": "Thanks for your work @alexteboul",
    "1679125": "with enough cropped samples, then  detector + classifier may work well."
  },
  "source": "meta"
}