{
  "id": 311967,
  "title": "Simple pipeline  ",
  "url": "/competitions/ultra-mnist/discussion/311967",
  "author_name": "Remek Kinas",
  "post_date": "2022-03-09T18:13:07.757000",
  "votes": 13,
  "comment_count": 0,
  "views": 0,
  "content": "<p>First simple processing pipeline which came to my mind:</p>\n<p><strong>1 Simplify - normalize background</strong> - <a href=\"https://www.kaggle.com/lukaszborecki/digit-cleaner-concept\" target=\"_blank\">Digit Cleaner Concept</a> from <a href=\"https://www.kaggle.com/lukaszborecki\" target=\"_blank\">@lukaszborecki</a> </p>\n<p><img src=\"https://i.ibb.co/0DZYLQG/ultra-mnist.jpg\" alt=\"Rd\"></p>\n<p><img src=\"https://i.ibb.co/sKxTJGB/ultra-mnist-001.jpg\" alt=\"Re\"></p>\n<p><img src=\"https://i.ibb.co/GWQD8Yj/ultra-mnist-002.jpg\" alt=\"Rt\"></p>\n<p>DATASET I CREATED -&gt; <a href=\"https://www.kaggle.com/remekkinas/ultramnistblack/\" target=\"_blank\">https://www.kaggle.com/remekkinas/ultramnistblack/</a></p>\n<p><strong>2 Detect numbers</strong></p>\n<p>Option 1 </p>\n<ul>\n<li>object detector (yolo 5 will be great) - 10 classes … (annotations is required)</li>\n<li>and SAHI - it should work perfectly in this case (detect on whole image and then on slices 4x4 - small numbers - build in feature) -&gt; <a href=\"https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx\" target=\"_blank\">https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx</a></li>\n</ul>\n<p>Option 2</p>\n<ul>\n<li>Computer Vision alghoritm to find regions with numbers</li>\n<li>Resize to 128x128</li>\n<li>Standard classification</li>\n</ul>\n<p>NOTEBOOK HERE: <a href=\"https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required\" target=\"_blank\">https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required</a></p>\n<p><strong>3 Sum numbers</strong> - deal with some problems - number interpretation (see my <a href=\"https://www.kaggle.com/remekkinas/funny-cv-eda-what-we-see-here\" target=\"_blank\">FUNNY CV EDA - what…. we see here …</a></p>\n<p><strong>4 Submit</strong></p>",
  "messages": [
    {
      "id": 1717216,
      "postDate": "2022-03-09T18:13:07.757Z",
      "content": "<p>First simple processing pipeline which came to my mind:</p>\n<p><strong>1 Simplify - normalize background</strong> - <a href=\"https://www.kaggle.com/lukaszborecki/digit-cleaner-concept\" target=\"_blank\">Digit Cleaner Concept</a> from <a href=\"https://www.kaggle.com/lukaszborecki\" target=\"_blank\">@lukaszborecki</a> </p>\n<p><img src=\"https://i.ibb.co/0DZYLQG/ultra-mnist.jpg\" alt=\"Rd\"></p>\n<p><img src=\"https://i.ibb.co/sKxTJGB/ultra-mnist-001.jpg\" alt=\"Re\"></p>\n<p><img src=\"https://i.ibb.co/GWQD8Yj/ultra-mnist-002.jpg\" alt=\"Rt\"></p>\n<p>DATASET I CREATED -&gt; <a href=\"https://www.kaggle.com/remekkinas/ultramnistblack/\" target=\"_blank\">https://www.kaggle.com/remekkinas/ultramnistblack/</a></p>\n<p><strong>2 Detect numbers</strong></p>\n<p>Option 1 </p>\n<ul>\n<li>object detector (yolo 5 will be great) - 10 classes … (annotations is required)</li>\n<li>and SAHI - it should work perfectly in this case (detect on whole image and then on slices 4x4 - small numbers - build in feature) -&gt; <a href=\"https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx\" target=\"_blank\">https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx</a></li>\n</ul>\n<p>Option 2</p>\n<ul>\n<li>Computer Vision alghoritm to find regions with numbers</li>\n<li>Resize to 128x128</li>\n<li>Standard classification</li>\n</ul>\n<p>NOTEBOOK HERE: <a href=\"https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required\" target=\"_blank\">https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required</a></p>\n<p><strong>3 Sum numbers</strong> - deal with some problems - number interpretation (see my <a href=\"https://www.kaggle.com/remekkinas/funny-cv-eda-what-we-see-here\" target=\"_blank\">FUNNY CV EDA - what…. we see here …</a></p>\n<p><strong>4 Submit</strong></p>",
      "rawMarkdown": "First simple processing pipeline which came to my mind:\n\n**1 Simplify - normalize background** - [Digit Cleaner Concept](https://www.kaggle.com/lukaszborecki/digit-cleaner-concept) from @lukaszborecki \n\n![Rd](https://i.ibb.co/0DZYLQG/ultra-mnist.jpg)\n\n![Re](https://i.ibb.co/sKxTJGB/ultra-mnist-001.jpg)\n\n![Rt](https://i.ibb.co/GWQD8Yj/ultra-mnist-002.jpg)\n\nDATASET I CREATED -> https://www.kaggle.com/remekkinas/ultramnistblack/\n\n**2 Detect numbers**\n\nOption 1 \n- object detector (yolo 5 will be great) - 10 classes ... (annotations is required)\n-  and SAHI - it should work perfectly in this case (detect on whole image and then on slices 4x4 - small numbers - build in feature) -> https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx\n\nOption 2\n- Computer Vision alghoritm to find regions with numbers\n- Resize to 128x128\n- Standard classification\n\nNOTEBOOK HERE: https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required\n\n**3 Sum numbers** - deal with some problems - number interpretation (see my [FUNNY CV EDA - what.... we see here ...](https://www.kaggle.com/remekkinas/funny-cv-eda-what-we-see-here)\n\n**4 Submit**",
      "votes": 13
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1717216": "First simple processing pipeline which came to my mind:\n\n**1 Simplify - normalize background** - [Digit Cleaner Concept](https://www.kaggle.com/lukaszborecki/digit-cleaner-concept) from @lukaszborecki \n\n![Rd](https://i.ibb.co/0DZYLQG/ultra-mnist.jpg)\n\n![Re](https://i.ibb.co/sKxTJGB/ultra-mnist-001.jpg)\n\n![Rt](https://i.ibb.co/GWQD8Yj/ultra-mnist-002.jpg)\n\nDATASET I CREATED -> https://www.kaggle.com/remekkinas/ultramnistblack/\n\n**2 Detect numbers**\n\nOption 1 \n- object detector (yolo 5 will be great) - 10 classes ... (annotations is required)\n-  and SAHI - it should work perfectly in this case (detect on whole image and then on slices 4x4 - small numbers - build in feature) -> https://www.kaggle.com/remekkinas/sahi-slicing-aided-hyper-inference-yv5-and-yx\n\nOption 2\n- Computer Vision alghoritm to find regions with numbers\n- Resize to 128x128\n- Standard classification\n\nNOTEBOOK HERE: https://www.kaggle.com/remekkinas/step-2-find-numbers-no-model-required\n\n**3 Sum numbers** - deal with some problems - number interpretation (see my [FUNNY CV EDA - what.... we see here ...](https://www.kaggle.com/remekkinas/funny-cv-eda-what-we-see-here)\n\n**4 Submit**"
  }
}