![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1984447%2F2e8f9bbbeb4a5ecbfbf152220c3eef89%2F3544220328_2.jpg?generation=1610826770227130&alt=media\")
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1984447%2Ff66c0580144f7a607c706929bdec6781%2F3545118265_1.jpg?generation=1610826779781918&alt=media\")
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1984447%2Ff7f197095714917c832886332f187bc6%2F3546224869_1.jpg?generation=1610826836458888&alt=media\")
I was using the below method to augment the image, but there are many black blocks in images.
\n---library albumentations---
\nreturn Compose([
\n RandomResizedCrop(512, 512),
\n Transpose(p=0.5),
\n HorizontalFlip(p=0.5),
\n VerticalFlip(p=0.5),
\n ShiftScaleRotate(p=0.5),
\n HueSaturationValue(hue_shift_limit=0.2, sat_shift_limit=0.2, val_shift_limit=0.2, p=0.5),
\n RandomBrightnessContrast(brightness_limit=(-0.1,0.1), contrast_limit=(-0.1, 0.1), p=0.5),
\n CoarseDropout(p=0.5),
\n ToTensorV2(),
\n ])
\n
\n
Need help about what leads this and how to deal with this. Is 'albumentations' not stable?
", "messages": [ { "id": "1155955", "postDate": "01/16/2021 19:57:11", "content": "I was using the below method to augment the image, but there are many black blocks in images.
\n---library albumentations---
\nreturn Compose([
\n RandomResizedCrop(512, 512),
\n Transpose(p=0.5),
\n HorizontalFlip(p=0.5),
\n VerticalFlip(p=0.5),
\n ShiftScaleRotate(p=0.5),
\n HueSaturationValue(hue_shift_limit=0.2, sat_shift_limit=0.2, val_shift_limit=0.2, p=0.5),
\n RandomBrightnessContrast(brightness_limit=(-0.1,0.1), contrast_limit=(-0.1, 0.1), p=0.5),
\n CoarseDropout(p=0.5),
\n ToTensorV2(),
\n ])
\n
\n
Need help about what leads this and how to deal with this. Is 'albumentations' not stable?
", "rawMarkdown": "I was using the below method to augment the image, but there are many black blocks in images.\n\n---library albumentations---\n\nreturn Compose([\n RandomResizedCrop(512, 512),\n Transpose(p=0.5),\n HorizontalFlip(p=0.5),\n VerticalFlip(p=0.5),\n ShiftScaleRotate(p=0.5),\n HueSaturationValue(hue_shift_limit=0.2, sat_shift_limit=0.2, val_shift_limit=0.2, p=0.5),\n RandomBrightnessContrast(brightness_limit=(-0.1,0.1), contrast_limit=(-0.1, 0.1), p=0.5),\n CoarseDropout(p=0.5),\n ToTensorV2(),\n ])\n\n\n\n\n\nNeed help about what leads this and how to deal with this. Is 'albumentations' not stable?", "votes": null }, { "id": "1155978", "postDate": "01/16/2021 20:16:33", "content": "that's because of CoarseDropout augmentation that you are using,read this paper : https://arxiv.org/abs/1708.04552
\nthose black points are part of your augmentation
Thanks, good to know.
", "rawMarkdown": "Thanks, good to know.", "votes": null }, { "id": "1156602", "postDate": "01/17/2021 09:28:02", "content": "Read the details on coarse dropout - you can change the number and size of the drop out areas - you need to make pretty large changes or have a very good cv method to find the best. Probably NOT a area to spend your compute on unless you have lots of compute resources - so just use the defaults.
", "rawMarkdown": "Read the details on coarse dropout - you can change the number and size of the drop out areas - you need to make pretty large changes or have a very good cv method to find the best. Probably NOT a area to spend your compute on unless you have lots of compute resources - so just use the defaults.", "votes": null } ], "comments": [ { "id": 1155978, "author_name": "mobassir", "author_url": "", "post_date": "01/16/2021 20:16:33", "content": "that's because of CoarseDropout augmentation that you are using,read this paper : https://arxiv.org/abs/1708.04552
\nthose black points are part of your augmentation
Thanks, good to know.
", "votes": null, "replies": [] } ] }, { "id": 1156602, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/17/2021 09:28:02", "content": "Read the details on coarse dropout - you can change the number and size of the drop out areas - you need to make pretty large changes or have a very good cv method to find the best. Probably NOT a area to spend your compute on unless you have lots of compute resources - so just use the defaults.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1155955": "I was using the below method to augment the image, but there are many black blocks in images.\n\n---library albumentations---\n\nreturn Compose([\n RandomResizedCrop(512, 512),\n Transpose(p=0.5),\n HorizontalFlip(p=0.5),\n VerticalFlip(p=0.5),\n ShiftScaleRotate(p=0.5),\n HueSaturationValue(hue_shift_limit=0.2, sat_shift_limit=0.2, val_shift_limit=0.2, p=0.5),\n RandomBrightnessContrast(brightness_limit=(-0.1,0.1), contrast_limit=(-0.1, 0.1), p=0.5),\n CoarseDropout(p=0.5),\n ToTensorV2(),\n ])\n\n\n\n\n\nNeed help about what leads this and how to deal with this. Is 'albumentations' not stable?", "1155978": "that's because of CoarseDropout augmentation that you are using,read this paper : https://arxiv.org/abs/1708.04552\nthose black points are part of your augmentation", "1156254": "Thanks, good to know.", "1156602": "Read the details on coarse dropout - you can change the number and size of the drop out areas - you need to make pretty large changes or have a very good cv method to find the best. Probably NOT a area to spend your compute on unless you have lots of compute resources - so just use the defaults." }, "source": "meta" }