{
  "id": 361561,
  "title": "What methods works best for Domain Adaptation (Cross-Dataset Generalization) tasks?",
  "url": "/competitions/hubmap-organ-segmentation/discussion/361561",
  "author_name": "ZavodRobotov",
  "post_date": "2022-10-22T10:15:05.156000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>This competition is mainly focuses on domain adaptation task. From top solutions I read, that Stain Normalization is good for such tasks. What other methods works best for Domain Adaptation (Cross-Dataset Generalization)  in you'r tasks?</p>\n<p>For me it is: </p>\n<ul>\n<li>Add more different datasets to training. This datasets can be not similar to target dataset, because thay whatever increase invariance to dataset (generalization beyond datasets).</li>\n<li>Histogram equalization</li>\n<li>Blend histogram equalization and CLAHE</li>\n<li>Color and contrast augmentations</li>\n<li>Small batch size</li>\n</ul>",
  "messages": [
    {
      "id": 1999413,
      "postDate": "2022-10-22T10:15:05.157Z",
      "content": "<p>This competition is mainly focuses on domain adaptation task. From top solutions I read, that Stain Normalization is good for such tasks. What other methods works best for Domain Adaptation (Cross-Dataset Generalization)  in you'r tasks?</p>\n<p>For me it is: </p>\n<ul>\n<li>Add more different datasets to training. This datasets can be not similar to target dataset, because thay whatever increase invariance to dataset (generalization beyond datasets).</li>\n<li>Histogram equalization</li>\n<li>Blend histogram equalization and CLAHE</li>\n<li>Color and contrast augmentations</li>\n<li>Small batch size</li>\n</ul>",
      "rawMarkdown": "This competition is mainly focuses on domain adaptation task. From top solutions I read, that Stain Normalization is good for such tasks. What other methods works best for Domain Adaptation (Cross-Dataset Generalization)  in you'r tasks?\n\nFor me it is: \n - Add more different datasets to training. This datasets can be not similar to target dataset, because thay whatever increase invariance to dataset (generalization beyond datasets).\n - Histogram equalization\n - Blend histogram equalization and CLAHE\n - Color and contrast augmentations\n - Small batch size",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1999413": "This competition is mainly focuses on domain adaptation task. From top solutions I read, that Stain Normalization is good for such tasks. What other methods works best for Domain Adaptation (Cross-Dataset Generalization)  in you'r tasks?\n\nFor me it is: \n - Add more different datasets to training. This datasets can be not similar to target dataset, because thay whatever increase invariance to dataset (generalization beyond datasets).\n - Histogram equalization\n - Blend histogram equalization and CLAHE\n - Color and contrast augmentations\n - Small batch size"
  }
}