{ "id": 284283, "title": "Lets talk Augmentations!", "url": "/competitions/sartorius-cell-instance-segmentation/discussion/284283", "author_name": "", "post_date": "2021-10-30T22:02:19.164000", "votes": 28, "comment_count": 3, "views": 0, "content": "

As you probably know, augmentations have repeatedly been shown to improve model accuracy.
\nHowever, so far I nearly didn't see any discussion or notebooks trying a heavy augmentations pipeline. so, let's talk about some augmentations you might want to try [some are not 100% relevant but are cool to include in a full summary]:
\nThis is a list of some great augmentations techniques used in the pas image competitions I participated in SIIM-ISIC & Bengali Grapheme Classification.
\nAugmentations
\n*Basic augmentation includes: flipping, rotation, sheer, zoom, and shift. Starter notebook here. Next, we can incorporate non-linear transformations like grid distortion and we can do color augmentation like brightness, contrast, hue, etc.
\n*Rotation Augmentation*
\n\"\"
\nFor each pixel in your augmented image, you must find which pixel value in the original image to use. In the example above, we wish to determine which pixel to place in location (1, 3). So we must multiply the coordinate (1, 3) by a rotation matrix to determine that we want pixel (3, 2) from the original image which is color pink. We then place a pink pixel in the destination image's pixel (1, 3).
\nCutMix and MixUp
\n\"\"
\nMixUp
\nMixUp blends two images and their one-hot encoded labels to create a new image. This helps a network generalize to the space of images they lie \"between\" the existing training images.
\n\"\"
\nAuto Augment
\nList a variety of augmentation techniques. Use reinforcement learning to explore different combinations of augmentations and find the best.
\n\"\"
\nCam CutMix
\nThis is cutout but we fill the blank rectangular space with a portion of another image. CutMix has the benefits of cutout (described above) and it expands the the space of images by creating new images (i.e. like a dog with a cat's head)
\n\"\"
\nAugMix
\nRandomly blend multiple images that have each been augmented themselves separately. And use a special consistency loss.
\n\"\"
\nCoarse Dropout and Cutout Augmentation GPU/TPU
\n\"\"
\nGridMask
\n\"\"

\n", "messages": [ { "id": 1565712, "postDate": "2021-10-30T22:02:19.163Z", "content": "

As you probably know, augmentations have repeatedly been shown to improve model accuracy.
\nHowever, so far I nearly didn't see any discussion or notebooks trying a heavy augmentations pipeline. so, let's talk about some augmentations you might want to try [some are not 100% relevant but are cool to include in a full summary]:
\nThis is a list of some great augmentations techniques used in the pas image competitions I participated in SIIM-ISIC & Bengali Grapheme Classification.
\nAugmentations
\n*Basic augmentation includes: flipping, rotation, sheer, zoom, and shift. Starter notebook here. Next, we can incorporate non-linear transformations like grid distortion and we can do color augmentation like brightness, contrast, hue, etc.
\n*Rotation Augmentation*
\n\"\"
\nFor each pixel in your augmented image, you must find which pixel value in the original image to use. In the example above, we wish to determine which pixel to place in location (1, 3). So we must multiply the coordinate (1, 3) by a rotation matrix to determine that we want pixel (3, 2) from the original image which is color pink. We then place a pink pixel in the destination image's pixel (1, 3).
\nCutMix and MixUp
\n\"\"
\nMixUp
\nMixUp blends two images and their one-hot encoded labels to create a new image. This helps a network generalize to the space of images they lie \"between\" the existing training images.
\n\"\"
\nAuto Augment
\nList a variety of augmentation techniques. Use reinforcement learning to explore different combinations of augmentations and find the best.
\n\"\"
\nCam CutMix
\nThis is cutout but we fill the blank rectangular space with a portion of another image. CutMix has the benefits of cutout (described above) and it expands the the space of images by creating new images (i.e. like a dog with a cat's head)
\n\"\"
\nAugMix
\nRandomly blend multiple images that have each been augmented themselves separately. And use a special consistency loss.
\n\"\"
\nCoarse Dropout and Cutout Augmentation GPU/TPU
\n\"\"
\nGridMask
\n\"\"

\n", "rawMarkdown": "\nAs you probably know, augmentations have repeatedly been shown to improve model accuracy.\nHowever, so far I nearly didn't see any discussion or notebooks trying a heavy augmentations pipeline. so, let's talk about some augmentations you might want to try [some are not 100% relevant but are cool to include in a full summary]: \n\nThis is a list of some great augmentations techniques used in the pas image competitions I participated in SIIM-ISIC & Bengali Grapheme Classification.\n\n**Augmentations**\n\n**Basic augmentation includes:* flipping, rotation, sheer, zoom, and shift. Starter notebook here. Next, we can incorporate non-linear transformations like grid distortion and we can do color augmentation like brightness, contrast, hue, etc.\n\n**Rotation Augmentation**\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fac5072f04b8e627020c439ab8326b038%2Frotate.JPG?generation=1582409737942526&alt=media)\n\nFor each pixel in your augmented image, you must find which pixel value in the original image to use. In the example above, we wish to determine which pixel to place in location (1, 3). So we must multiply the coordinate (1, 3) by a rotation matrix to determine that we want pixel (3, 2) from the original image which is color pink. We then place a pink pixel in the destination image's pixel (1, 3).\n\n**CutMix and MixUp**\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F999713370831a44bb5af627f96b42f76%2FScreen%20Shot%202020-02-28%20at%2012.09.49%20PM.png?generation=1582920604632237&alt=media)\n\n**MixUp**\n\nMixUp blends two images and their one-hot encoded labels to create a new image. This helps a network generalize to the space of images they lie \"between\" the existing training images.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F0fd1c6f6c0d1b2c220748e1e8517dd42%2FScreen%20Shot%202020-02-28%20at%2012.04.14%20PM.png?generation=1582920270999369&alt=media)\n\n**Auto Augment**\n\nList a variety of augmentation techniques. Use reinforcement learning to explore different combinations of augmentations and find the best.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fa013e63820a9f2e5751fc954bd502561%2FScreen%20Shot%202020-02-28%20at%2012.05.08%20PM.png?generation=1582920325192602&alt=media)\n\n**Cam CutMix**\n\nThis is cutout but we fill the blank rectangular space with a portion of another image. CutMix has the benefits of cutout (described above) and it expands the the space of images by creating new images (i.e. like a dog with a cat's head)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fa4d6d7b6132350ac39dbe5d3ce601b55%2FScreen%20Shot%202020-02-28%20at%2012.06.24%20PM.png?generation=1582920401462775&alt=media)\n\n**AugMix**\n\nRandomly blend multiple images that have each been augmented themselves separately. And use a special consistency loss.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F2b79e1180f717408930554343b007881%2FScreen%20Shot%202020-02-28%20at%2012.07.34%20PM.png?generation=1582920469116799&alt=media)\n\n**Coarse Dropout and Cutout Augmentation GPU/TPU**\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F8db206436734b5f0c7b2179c2e152eb0%2FScreen%20Shot%202020-07-22%20at%201.28.49%20PM.png?generation=1595632223386869&alt=media)\n\n\n**GridMask**\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3617078%2F5bd38f98a37535a92970b5b338e519f6%2F2020-02-29%201.32.04.png?generation=1582954461136321&alt=media)\n\n\n- Reference\n\n- Credit to the original post on SIIM ISIC, Heavily based on it: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/160430\n\n- [How To - Rotation Augmentation GPU/TPU by Chris Deotte](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132191)\n\n- [How To - CutMix and MixUp on GPU/TPU by Chris Deotte](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132935)\n\n- [GridMask data augmentation on GPU/TPU](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132986)\n\n- [Microscope augmentation](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159476)\n\n- [Advanced hair augmentation](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159176)\n\n- [CAM Cutmix](https://www.kaggle.com/c/bengaliai-cv19/discussion/136025)\n\n- [Coarse Dropout and Cutout Augmentation GPU/TPU by Chris Deotte](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/169721)\n\n", "votes": 28 }, { "id": 1584446, "postDate": "2021-11-16T14:35:55.730Z", "content": "

Thank you for sharing your experience!
\nI am not familiar with data augmentation.
\nThis paper gave me the overview of it:)
\nhttps://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0

", "rawMarkdown": "Thank you for sharing your experience!\nI am not familiar with data augmentation.\nThis paper gave me the overview of it:)\nhttps://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0", "votes": 1 }, { "id": 1583275, "postDate": "2021-11-15T17:19:33.420Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1565857, "postDate": "2021-10-31T05:36:16.007Z", "content": "

Thanks for your sharing.

", "rawMarkdown": "Thanks for your sharing. " } ], "comments": [ { "id": 1584446, "author_name": "John Doe", "author_url": "", "post_date": "2021-11-16T14:35:55.730000", "content": "

Thank you for sharing your experience!
\nI am not familiar with data augmentation.
\nThis paper gave me the overview of it:)
\nhttps://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0

", "votes": 1, "replies": [] }, { "id": 1583275, "author_name": "", "author_url": "", "post_date": "2021-11-15T17:19:33.420000", "content": "", "votes": 0, "replies": [] }, { "id": 1565857, "author_name": "Benny Fung", "author_url": "", "post_date": "2021-10-31T05:36:16.007000", "content": "

Thanks for your sharing.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1565712": "\nAs you probably know, augmentations have repeatedly been shown to improve model accuracy.\nHowever, so far I nearly didn't see any discussion or notebooks trying a heavy augmentations pipeline. so, let's talk about some augmentations you might want to try [some are not 100% relevant but are cool to include in a full summary]: \n\nThis is a list of some great augmentations techniques used in the pas image competitions I participated in SIIM-ISIC & Bengali Grapheme Classification.\n\n**Augmentations**\n\n**Basic augmentation includes:* flipping, rotation, sheer, zoom, and shift. Starter notebook here. Next, we can incorporate non-linear transformations like grid distortion and we can do color augmentation like brightness, contrast, hue, etc.\n\n**Rotation Augmentation**\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fac5072f04b8e627020c439ab8326b038%2Frotate.JPG?generation=1582409737942526&alt=media)\n\nFor each pixel in your augmented image, you must find which pixel value in the original image to use. In the example above, we wish to determine which pixel to place in location (1, 3). So we must multiply the coordinate (1, 3) by a rotation matrix to determine that we want pixel (3, 2) from the original image which is color pink. We then place a pink pixel in the destination image's pixel (1, 3).\n\n**CutMix and MixUp**\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F999713370831a44bb5af627f96b42f76%2FScreen%20Shot%202020-02-28%20at%2012.09.49%20PM.png?generation=1582920604632237&alt=media)\n\n**MixUp**\n\nMixUp blends two images and their one-hot encoded labels to create a new image. This helps a network generalize to the space of images they lie \"between\" the existing training images.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F0fd1c6f6c0d1b2c220748e1e8517dd42%2FScreen%20Shot%202020-02-28%20at%2012.04.14%20PM.png?generation=1582920270999369&alt=media)\n\n**Auto Augment**\n\nList a variety of augmentation techniques. Use reinforcement learning to explore different combinations of augmentations and find the best.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fa013e63820a9f2e5751fc954bd502561%2FScreen%20Shot%202020-02-28%20at%2012.05.08%20PM.png?generation=1582920325192602&alt=media)\n\n**Cam CutMix**\n\nThis is cutout but we fill the blank rectangular space with a portion of another image. CutMix has the benefits of cutout (described above) and it expands the the space of images by creating new images (i.e. like a dog with a cat's head)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2Fa4d6d7b6132350ac39dbe5d3ce601b55%2FScreen%20Shot%202020-02-28%20at%2012.06.24%20PM.png?generation=1582920401462775&alt=media)\n\n**AugMix**\n\nRandomly blend multiple images that have each been augmented themselves separately. And use a special consistency loss.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F2b79e1180f717408930554343b007881%2FScreen%20Shot%202020-02-28%20at%2012.07.34%20PM.png?generation=1582920469116799&alt=media)\n\n**Coarse Dropout and Cutout Augmentation GPU/TPU**\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1723677%2F8db206436734b5f0c7b2179c2e152eb0%2FScreen%20Shot%202020-07-22%20at%201.28.49%20PM.png?generation=1595632223386869&alt=media)\n\n\n**GridMask**\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3617078%2F5bd38f98a37535a92970b5b338e519f6%2F2020-02-29%201.32.04.png?generation=1582954461136321&alt=media)\n\n\n- Reference\n\n- Credit to the original post on SIIM ISIC, Heavily based on it: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/160430\n\n- [How To - Rotation Augmentation GPU/TPU by Chris Deotte](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132191)\n\n- [How To - CutMix and MixUp on GPU/TPU by Chris Deotte](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132935)\n\n- [GridMask data augmentation on GPU/TPU](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/132986)\n\n- [Microscope augmentation](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159476)\n\n- [Advanced hair augmentation](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159176)\n\n- [CAM Cutmix](https://www.kaggle.com/c/bengaliai-cv19/discussion/136025)\n\n- [Coarse Dropout and Cutout Augmentation GPU/TPU by Chris Deotte](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/169721)\n\n", "1584446": "Thank you for sharing your experience!\nI am not familiar with data augmentation.\nThis paper gave me the overview of it:)\nhttps://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0", "1583275": "", "1565857": "Thanks for your sharing. " } }