{
  "id": 53145,
  "title": "what is TTA",
  "url": "/competitions/sp-society-camera-model-identification/discussion/53145",
  "author_name": "",
  "post_date": "2018-03-27T16:43:29.581130700Z",
  "votes": 7,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I am a novice to image processing~ I see many methods using TTA. But I couldn't find definition of TTA by Google. Anyone can share me some material about TTA?  thanks!!!</p>",
  "messages": [
    {
      "id": "304455",
      "postDate": "03/27/2018 16:43:29",
      "content": "<p>I am a novice to image processing~ I see many methods using TTA. But I couldn't find definition of TTA by Google. Anyone can share me some material about TTA?  thanks!!!</p>",
      "rawMarkdown": "I am a novice to image processing~ I see many methods using TTA. But I couldn't find definition of TTA by Google. Anyone can share me some material about TTA?  thanks!!!",
      "votes": null
    },
    {
      "id": "304481",
      "postDate": "03/27/2018 17:08:07",
      "content": "<p><a href=\"https://github.com/fastai/fastai/blob/0b09e4e7f347b2c7930a123a6facb9a397329211/fastai/learner.py\">TTA</a> stands for Test Time Augmentation and is described in the following blogpost: <a href=\"http://benanne.github.io/2015/03/17/plankton.html\">http://benanne.github.io/2015/03/17/plankton.html</a></p>\n\n<pre><code>def TTA(self, n_aug=4, is_test=False):\n        \"\"\" Predict with Test Time Augmentation (TTA)\n        Additional to the original test/validation images, apply image augmentation to them\n        (just like for training images) and calculate the mean of predictions. The intent\n        is to increase the accuracy of predictions by examining the images using multiple\n        perspectives.\n        Args:\n            n_aug: a number of augmentation images to use per original image\n            is_test: indicate to use test images; otherwise use validation images\n        Returns:\n            (tuple): a tuple containing:\n                log predictions (numpy.ndarray): log predictions (i.e. `np.exp(log_preds)` will return probabilities)\n                targs (numpy.ndarray): target values when `is_test==False`; zeros otherwise.\n        \"\"\"\n</code></pre>\n\n<p>Source: <a href=\"https://github.com/fastai\">fastai</a></p>",
      "rawMarkdown": "[TTA][1] stands for Test Time Augmentation and is described in the following blogpost: http://benanne.github.io/2015/03/17/plankton.html\n\n    def TTA(self, n_aug=4, is_test=False):\n            \"\"\" Predict with Test Time Augmentation (TTA)\n            Additional to the original test/validation images, apply image augmentation to them\n            (just like for training images) and calculate the mean of predictions. The intent\n            is to increase the accuracy of predictions by examining the images using multiple\n            perspectives.\n            Args:\n                n_aug: a number of augmentation images to use per original image\n                is_test: indicate to use test images; otherwise use validation images\n            Returns:\n                (tuple): a tuple containing:\n                    log predictions (numpy.ndarray): log predictions (i.e. `np.exp(log_preds)` will return probabilities)\n                    targs (numpy.ndarray): target values when `is_test==False`; zeros otherwise.\n            \"\"\"\nSource: [fastai][2]\n\n\n\n  [1]: https://github.com/fastai/fastai/blob/0b09e4e7f347b2c7930a123a6facb9a397329211/fastai/learner.py\n  [2]: https://github.com/fastai",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 304481,
      "author_name": "paultimothymooney",
      "author_url": "",
      "post_date": "03/27/2018 17:08:07",
      "content": "<p><a href=\"https://github.com/fastai/fastai/blob/0b09e4e7f347b2c7930a123a6facb9a397329211/fastai/learner.py\">TTA</a> stands for Test Time Augmentation and is described in the following blogpost: <a href=\"http://benanne.github.io/2015/03/17/plankton.html\">http://benanne.github.io/2015/03/17/plankton.html</a></p>\n\n<pre><code>def TTA(self, n_aug=4, is_test=False):\n        \"\"\" Predict with Test Time Augmentation (TTA)\n        Additional to the original test/validation images, apply image augmentation to them\n        (just like for training images) and calculate the mean of predictions. The intent\n        is to increase the accuracy of predictions by examining the images using multiple\n        perspectives.\n        Args:\n            n_aug: a number of augmentation images to use per original image\n            is_test: indicate to use test images; otherwise use validation images\n        Returns:\n            (tuple): a tuple containing:\n                log predictions (numpy.ndarray): log predictions (i.e. `np.exp(log_preds)` will return probabilities)\n                targs (numpy.ndarray): target values when `is_test==False`; zeros otherwise.\n        \"\"\"\n</code></pre>\n\n<p>Source: <a href=\"https://github.com/fastai\">fastai</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "304455": "I am a novice to image processing~ I see many methods using TTA. But I couldn't find definition of TTA by Google. Anyone can share me some material about TTA?  thanks!!!",
    "304481": "[TTA][1] stands for Test Time Augmentation and is described in the following blogpost: http://benanne.github.io/2015/03/17/plankton.html\n\n    def TTA(self, n_aug=4, is_test=False):\n            \"\"\" Predict with Test Time Augmentation (TTA)\n            Additional to the original test/validation images, apply image augmentation to them\n            (just like for training images) and calculate the mean of predictions. The intent\n            is to increase the accuracy of predictions by examining the images using multiple\n            perspectives.\n            Args:\n                n_aug: a number of augmentation images to use per original image\n                is_test: indicate to use test images; otherwise use validation images\n            Returns:\n                (tuple): a tuple containing:\n                    log predictions (numpy.ndarray): log predictions (i.e. `np.exp(log_preds)` will return probabilities)\n                    targs (numpy.ndarray): target values when `is_test==False`; zeros otherwise.\n            \"\"\"\nSource: [fastai][2]\n\n\n\n  [1]: https://github.com/fastai/fastai/blob/0b09e4e7f347b2c7930a123a6facb9a397329211/fastai/learner.py\n  [2]: https://github.com/fastai"
  },
  "source": "meta"
}