![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2F2184811fc18555da73a64277ab5016a1%2F2020-01-30%2011.33.43.png?generation=1580351647542803&alt=media\")
https://arxiv.org/abs/2001.04086
\n\n
https://www.kaggle.com/haqishen/gridmask
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2F0916989aab4b5915c6d052893b125f00%2F1.png?generation=1580292219766034&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2Fc4e1c0ad186ddb522e80d95f3c92dcee%2F2.png?generation=1580292232578384&alt=media\")
https://arxiv.org/abs/2001.04086
\n\n
https://www.kaggle.com/haqishen/gridmask
\n\n
Thanks for your sharing!!!
", "rawMarkdown": "Thanks for your sharing!!!", "votes": null }, { "id": "732059", "postDate": "01/29/2020 12:41:44", "content": "Excellent!!Thank you!!
", "rawMarkdown": "Excellent!!Thank you!!", "votes": null }, { "id": "732203", "postDate": "01/29/2020 15:10:08", "content": "Nice work!!!
", "rawMarkdown": "Nice work!!!", "votes": null }, { "id": "732413", "postDate": "01/29/2020 19:39:12", "content": "Excellent👍
", "rawMarkdown": "Excellent👍", "votes": null }, { "id": "732741", "postDate": "01/30/2020 07:31:26", "content": "Good job 💯
", "rawMarkdown": "Good job 💯", "votes": null }, { "id": "733003", "postDate": "01/30/2020 14:49:50", "content": "This is like sprinkles: https://medium.com/@lessw/progressive-sprinkles-a-new-data-augmentation-for-cnns-and-helps-achieve-new-98-nih-malaria-6056965f671a\nthanks for sharing though
", "rawMarkdown": "This is like sprinkles: https://medium.com/@lessw/progressive-sprinkles-a-new-data-augmentation-for-cnns-and-helps-achieve-new-98-nih-malaria-6056965f671a\nthanks for sharing though", "votes": null }, { "id": "734364", "postDate": "02/01/2020 11:09:49", "content": "Thank you for your sharing!
", "rawMarkdown": "Thank you for your sharing!", "votes": null }, { "id": "737111", "postDate": "02/04/2020 23:35:01", "content": "Thanks for sharing @haqishen !
", "rawMarkdown": "Thanks for sharing @haqishen !", "votes": null }, { "id": "737511", "postDate": "02/05/2020 12:56:59", "content": "i'm trying gridmask but from my experiment gridmask converge super slow :( anyone facing the same problem?
", "rawMarkdown": "i'm trying gridmask but from my experiment gridmask converge super slow :( anyone facing the same problem?", "votes": null }, { "id": "737599", "postDate": "02/05/2020 15:03:49", "content": "My code works fine for me.\nCould you attach some images that applied GridMask here?
", "rawMarkdown": "My code works fine for me.\nCould you attach some images that applied GridMask here?", "votes": null }, { "id": "737615", "postDate": "02/05/2020 15:22:19", "content": "my images look fine - your implement is great :D but maybe i'm too greedy by setting big parameters. for same recall score, i can see the training loss is around 3x-4x compared to without gridmask. big regularizer will call for longer training epoches
", "rawMarkdown": "my images look fine - your implement is great :D but maybe i'm too greedy by setting big parameters. for same recall score, i can see the training loss is around 3x-4x compared to without gridmask. big regularizer will call for longer training epoches", "votes": null }, { "id": "738265", "postDate": "02/06/2020 10:39:23", "content": "Great work!
", "rawMarkdown": "Great work!", "votes": null }, { "id": "738731", "postDate": "02/06/2020 23:42:20", "content": "Thanks for sharing!!👍
", "rawMarkdown": "Thanks for sharing!!👍", "votes": null }, { "id": "741401", "postDate": "02/10/2020 15:33:40", "content": "After a series of experiments, I'm quite confident that with some specific parameter setting, GridMask can improve the model performance on this dataset.\nIt will surely be part of my final solution.
", "rawMarkdown": "After a series of experiments, I'm quite confident that with some specific parameter setting, GridMask can improve the model performance on this dataset.\nIt will surely be part of my final solution.", "votes": null }, { "id": "742177", "postDate": "02/11/2020 06:03:24", "content": "Thanks for the kernel. I will use it .
", "rawMarkdown": "Thanks for the kernel. I will use it .", "votes": null }, { "id": "742989", "postDate": "02/11/2020 17:23:44", "content": "@haqishen Did it give better results than mixup/cutmix for you ?
", "rawMarkdown": "haqishen Did it give better results than mixup/cutmix for you ?", "votes": null }, { "id": "744630", "postDate": "02/13/2020 02:18:43", "content": "I wonder if GridMask still work under mixup/cutmix.I did't make it work(model: se-resnext-50 augement with mixup(0.4)/cutmix(0.8-1) cv:0.971(Gridmask)/0.9737(shear rotate shift scale))) 😨
", "rawMarkdown": "I wonder if GridMask still work under mixup/cutmix.I did't make it work(model: se-resnext-50 augement with mixup(0.4)/cutmix(0.8-1) cv:0.971(Gridmask)/0.9737(shear rotate shift scale))) 😨", "votes": null }, { "id": "750012", "postDate": "02/19/2020 04:07:27", "content": "Can you explain what \"big parameters\" and \"big regularizer\" mean? Thanks!
", "rawMarkdown": "Can you explain what \"big parameters\" and \"big regularizer\" mean? Thanks!", "votes": null }, { "id": "785889", "postDate": "03/25/2020 13:50:34", "content": "Thx for your great work. BTW, I would like to ask how to create our own transform class in albumentations?\nSpecifically, I'm quite confused about functions such as 'get_params_dependent_on_targets', 'get_transform_init_args_names', etc. \nAnd I haven't found answer in albumentations documents.
", "rawMarkdown": "Thx for your great work. BTW, I would like to ask how to create our own transform class in albumentations?\nSpecifically, I'm quite confused about functions such as 'get_params_dependent_on_targets', 'get_transform_init_args_names', etc. \nAnd I haven't found answer in albumentations documents.", "votes": null }, { "id": "785927", "postDate": "03/25/2020 14:16:32", "content": "Ah... I found my answer in the source code.\nIn case someone else meets the same problem, they should read source code in transforms_interface.py
", "rawMarkdown": "Ah... I found my answer in the source code.\nIn case someone else meets the same problem, they should read source code in transforms_interface.py", "votes": null } ], "comments": [ { "id": 731973, "author_name": "hesene", "author_url": "", "post_date": "01/29/2020 10:21:42", "content": "Thanks for your sharing!!!
", "votes": null, "replies": [] }, { "id": 732059, "author_name": "changewow", "author_url": "", "post_date": "01/29/2020 12:41:44", "content": "Excellent!!Thank you!!
", "votes": null, "replies": [] }, { "id": 732203, "author_name": "victormenuzzo", "author_url": "", "post_date": "01/29/2020 15:10:08", "content": "Nice work!!!
", "votes": null, "replies": [] }, { "id": 732413, "author_name": "nitwmanish", "author_url": "", "post_date": "01/29/2020 19:39:12", "content": "Excellent👍
", "votes": null, "replies": [] }, { "id": 732741, "author_name": "dasmehdixtr", "author_url": "", "post_date": "01/30/2020 07:31:26", "content": "Good job 💯
", "votes": null, "replies": [] }, { "id": 733003, "author_name": "timetraveller98", "author_url": "", "post_date": "01/30/2020 14:49:50", "content": "This is like sprinkles: https://medium.com/@lessw/progressive-sprinkles-a-new-data-augmentation-for-cnns-and-helps-achieve-new-98-nih-malaria-6056965f671a\nthanks for sharing though
", "votes": null, "replies": [] }, { "id": 734364, "author_name": "liang23333", "author_url": "", "post_date": "02/01/2020 11:09:49", "content": "Thank you for your sharing!
", "votes": null, "replies": [] }, { "id": 737111, "author_name": "rohitagarwal", "author_url": "", "post_date": "02/04/2020 23:35:01", "content": "Thanks for sharing @haqishen !
", "votes": null, "replies": [] }, { "id": 737511, "author_name": "moewie94", "author_url": "", "post_date": "02/05/2020 12:56:59", "content": "i'm trying gridmask but from my experiment gridmask converge super slow :( anyone facing the same problem?
", "votes": null, "replies": [ { "id": 737599, "author_name": "haqishen", "author_url": "", "post_date": "02/05/2020 15:03:49", "content": "My code works fine for me.\nCould you attach some images that applied GridMask here?
", "votes": null, "replies": [] }, { "id": 737615, "author_name": "moewie94", "author_url": "", "post_date": "02/05/2020 15:22:19", "content": "my images look fine - your implement is great :D but maybe i'm too greedy by setting big parameters. for same recall score, i can see the training loss is around 3x-4x compared to without gridmask. big regularizer will call for longer training epoches
", "votes": null, "replies": [] }, { "id": 750012, "author_name": "zhangyang", "author_url": "", "post_date": "02/19/2020 04:07:27", "content": "Can you explain what \"big parameters\" and \"big regularizer\" mean? Thanks!
", "votes": null, "replies": [] } ] }, { "id": 738265, "author_name": "", "author_url": "", "post_date": "02/06/2020 10:39:23", "content": "Great work!
", "votes": null, "replies": [] }, { "id": 738731, "author_name": "yuanlin08", "author_url": "", "post_date": "02/06/2020 23:42:20", "content": "Thanks for sharing!!👍
", "votes": null, "replies": [] }, { "id": 741401, "author_name": "haqishen", "author_url": "", "post_date": "02/10/2020 15:33:40", "content": "After a series of experiments, I'm quite confident that with some specific parameter setting, GridMask can improve the model performance on this dataset.\nIt will surely be part of my final solution.
", "votes": null, "replies": [ { "id": 742989, "author_name": "vladvdv", "author_url": "", "post_date": "02/11/2020 17:23:44", "content": "@haqishen Did it give better results than mixup/cutmix for you ?
", "votes": null, "replies": [] } ] }, { "id": 742177, "author_name": "virajbagal", "author_url": "", "post_date": "02/11/2020 06:03:24", "content": "Thanks for the kernel. I will use it .
", "votes": null, "replies": [] }, { "id": 744630, "author_name": "a457435687", "author_url": "", "post_date": "02/13/2020 02:18:43", "content": "I wonder if GridMask still work under mixup/cutmix.I did't make it work(model: se-resnext-50 augement with mixup(0.4)/cutmix(0.8-1) cv:0.971(Gridmask)/0.9737(shear rotate shift scale))) 😨
", "votes": null, "replies": [] }, { "id": 785889, "author_name": "jeffwang987", "author_url": "", "post_date": "03/25/2020 13:50:34", "content": "Thx for your great work. BTW, I would like to ask how to create our own transform class in albumentations?\nSpecifically, I'm quite confused about functions such as 'get_params_dependent_on_targets', 'get_transform_init_args_names', etc. \nAnd I haven't found answer in albumentations documents.
", "votes": null, "replies": [ { "id": 785927, "author_name": "jeffwang987", "author_url": "", "post_date": "03/25/2020 14:16:32", "content": "Ah... I found my answer in the source code.\nIn case someone else meets the same problem, they should read source code in transforms_interface.py
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "731968": "## GridMask\n\nhttps://arxiv.org/abs/2001.04086\n\n\n\n\n## Implementation & Demo\n\nhttps://www.kaggle.com/haqishen/gridmask\n\n## Output\n\n\n\n", "731973": "Thanks for your sharing!!!", "732059": "Excellent!!Thank you!!", "732203": "Nice work!!!", "732413": "Excellent👍", "732741": "Good job 💯", "733003": "This is like sprinkles: https://medium.com/@lessw/progressive-sprinkles-a-new-data-augmentation-for-cnns-and-helps-achieve-new-98-nih-malaria-6056965f671a\nthanks for sharing though", "734364": "Thank you for your sharing!", "737111": "Thanks for sharing @haqishen !", "737511": "i'm trying gridmask but from my experiment gridmask converge super slow :( anyone facing the same problem?", "737599": "My code works fine for me.\nCould you attach some images that applied GridMask here?", "737615": "my images look fine - your implement is great :D but maybe i'm too greedy by setting big parameters. for same recall score, i can see the training loss is around 3x-4x compared to without gridmask. big regularizer will call for longer training epoches", "738265": "Great work!", "738731": "Thanks for sharing!!👍", "741401": "After a series of experiments, I'm quite confident that with some specific parameter setting, GridMask can improve the model performance on this dataset.\nIt will surely be part of my final solution.", "742177": "Thanks for the kernel. I will use it .", "742989": "haqishen Did it give better results than mixup/cutmix for you ?", "744630": "I wonder if GridMask still work under mixup/cutmix.I did't make it work(model: se-resnext-50 augement with mixup(0.4)/cutmix(0.8-1) cv:0.971(Gridmask)/0.9737(shear rotate shift scale))) 😨", "750012": "Can you explain what \"big parameters\" and \"big regularizer\" mean? Thanks!", "785889": "Thx for your great work. BTW, I would like to ask how to create our own transform class in albumentations?\nSpecifically, I'm quite confused about functions such as 'get_params_dependent_on_targets', 'get_transform_init_args_names', etc. \nAnd I haven't found answer in albumentations documents.", "785927": "Ah... I found my answer in the source code.\nIn case someone else meets the same problem, they should read source code in transforms_interface.py" }, "source": "meta" }