{
  "id": 134042,
  "title": "What is your strategy of tuning data augmentation parameters?",
  "url": "/competitions/bengaliai-cv19/discussion/134042",
  "author_name": "YS",
  "post_date": "2020-03-05T15:24:54.449000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I was wondering what is your strategy of tuning data augmentation parameters?\nI felt my strategy is bad and wasting time.</p>\n\n<p>I personally tuned data augmentation parameters by showing images and trained the augmented data with a model.</p>\n\n<p>For example, \na small experiment:\nbaseline 0.95\nbaseline + rotation (10 degrees) 0.953\nbaseline + Mixup (alpha=0.4)  0.958</p>\n\n<p>Start to train a model based on previous experiment:\nTrained a model with mixup (alpha=0.4) first, and get model w1.\nThen train the last model (w1) with rotation (10 degrees) . However, performance is worse than the last model performance.</p>\n\n<p>I thought my strategy of tuning data augmentation parameters is very bad.\nCould you share how you tuning data augmentation parameters?</p>",
  "messages": [
    {
      "id": 764544,
      "postDate": "2020-03-05T15:24:54.450Z",
      "content": "<p>I was wondering what is your strategy of tuning data augmentation parameters?\nI felt my strategy is bad and wasting time.</p>\n\n<p>I personally tuned data augmentation parameters by showing images and trained the augmented data with a model.</p>\n\n<p>For example, \na small experiment:\nbaseline 0.95\nbaseline + rotation (10 degrees) 0.953\nbaseline + Mixup (alpha=0.4)  0.958</p>\n\n<p>Start to train a model based on previous experiment:\nTrained a model with mixup (alpha=0.4) first, and get model w1.\nThen train the last model (w1) with rotation (10 degrees) . However, performance is worse than the last model performance.</p>\n\n<p>I thought my strategy of tuning data augmentation parameters is very bad.\nCould you share how you tuning data augmentation parameters?</p>",
      "rawMarkdown": "I was wondering what is your strategy of tuning data augmentation parameters?\nI felt my strategy is bad and wasting time.\n\nI personally tuned data augmentation parameters by showing images and trained the augmented data with a model.\n\nFor example, \na small experiment:\nbaseline 0.95\nbaseline + rotation (10 degrees) 0.953\nbaseline + Mixup (alpha=0.4)  0.958\n\nStart to train a model based on previous experiment:\nTrained a model with mixup (alpha=0.4) first, and get model w1.\nThen train the last model (w1) with rotation (10 degrees) . However, performance is worse than the last model performance.\n\nI thought my strategy of tuning data augmentation parameters is very bad.\nCould you share how you tuning data augmentation parameters?",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "764544": "I was wondering what is your strategy of tuning data augmentation parameters?\nI felt my strategy is bad and wasting time.\n\nI personally tuned data augmentation parameters by showing images and trained the augmented data with a model.\n\nFor example, \na small experiment:\nbaseline 0.95\nbaseline + rotation (10 degrees) 0.953\nbaseline + Mixup (alpha=0.4)  0.958\n\nStart to train a model based on previous experiment:\nTrained a model with mixup (alpha=0.4) first, and get model w1.\nThen train the last model (w1) with rotation (10 degrees) . However, performance is worse than the last model performance.\n\nI thought my strategy of tuning data augmentation parameters is very bad.\nCould you share how you tuning data augmentation parameters?"
  }
}