{
  "id": 475060,
  "title": "132th Solution: 2D UNet with p1-p99 normalization",
  "url": "/competitions/blood-vessel-segmentation/discussion/475060",
  "author_name": "Ángel Jacinto Sánchez Ruiz",
  "post_date": "2024-02-07T01:19:53.900000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>A simple tu-seresnext50_32x4d UNet with volumes normalized on percentiles 1 and 99. The inference was a votation accumulation from x3 axis x4 rot90 scan of the minimum square possible padding with reflections. TH = 6 (of 12, .5)<br>\n<a href=\"https://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508\" target=\"_blank\">https://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508</a> <br>\nThe training was made on kidney_1_dense and kidney_3_dense 512x512 centred on labels centroid crops padded with reflections too and basic augmentations. Batches of 8 (4 of each kidney) during 50 epochs without validation:<br>\n<a href=\"https://www.kaggle.com/code/sacuscreed/143th-solution\" target=\"_blank\">https://www.kaggle.com/code/sacuscreed/143th-solution</a></p>",
  "messages": [
    {
      "id": 2640546,
      "postDate": "2024-02-07T01:19:53.900Z",
      "content": "<p>A simple tu-seresnext50_32x4d UNet with volumes normalized on percentiles 1 and 99. The inference was a votation accumulation from x3 axis x4 rot90 scan of the minimum square possible padding with reflections. TH = 6 (of 12, .5)<br>\n<a href=\"https://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508\" target=\"_blank\">https://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508</a> <br>\nThe training was made on kidney_1_dense and kidney_3_dense 512x512 centred on labels centroid crops padded with reflections too and basic augmentations. Batches of 8 (4 of each kidney) during 50 epochs without validation:<br>\n<a href=\"https://www.kaggle.com/code/sacuscreed/143th-solution\" target=\"_blank\">https://www.kaggle.com/code/sacuscreed/143th-solution</a></p>",
      "rawMarkdown": "A simple tu-seresnext50_32x4d UNet with volumes normalized on percentiles 1 and 99. The inference was a votation accumulation from x3 axis x4 rot90 scan of the minimum square possible padding with reflections. TH = 6 (of 12, .5)\nhttps://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508 \nThe training was made on kidney_1_dense and kidney_3_dense 512x512 centred on labels centroid crops padded with reflections too and basic augmentations. Batches of 8 (4 of each kidney) during 50 epochs without validation:\nhttps://www.kaggle.com/code/sacuscreed/143th-solution"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2640546": "A simple tu-seresnext50_32x4d UNet with volumes normalized on percentiles 1 and 99. The inference was a votation accumulation from x3 axis x4 rot90 scan of the minimum square possible padding with reflections. TH = 6 (of 12, .5)\nhttps://www.kaggle.com/code/sacuscreed/smp-2d-inference-back-to-basics?scriptVersionId=161523508 \nThe training was made on kidney_1_dense and kidney_3_dense 512x512 centred on labels centroid crops padded with reflections too and basic augmentations. Batches of 8 (4 of each kidney) during 50 epochs without validation:\nhttps://www.kaggle.com/code/sacuscreed/143th-solution"
  }
}