{ "id": 212060, "title": "Try Mixup Without Hesitation", "url": "/competitions/cassava-leaf-disease-classification/discussion/212060", "author_name": "", "post_date": "2021-01-17T10:23:33.763505900Z", "votes": 21, "comment_count": 5, "views": 0, "content": "

\"\"

\n

paper : https://arxiv.org/abs/2101.04342

\n

code : https://github.com/yuhao318/mwh

", "messages": [ { "id": "1156676", "postDate": "01/17/2021 10:23:33", "content": "

\"\"

\n

paper : https://arxiv.org/abs/2101.04342

\n

code : https://github.com/yuhao318/mwh

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2034058%2F9640da7324cea7d6953221bd443259c8%2Fmix.jpg?generation=1610878944120116&alt=media)\n\n\n\n\n# paper : https://arxiv.org/abs/2101.04342\n\n# code : https://github.com/yuhao318/mwh", "votes": null }, { "id": "1157780", "postDate": "01/18/2021 06:22:38", "content": "

Hello, in my experiment, Mixup didn't improve my score both local and LB. Does it works for you?

", "rawMarkdown": "Hello, in my experiment, Mixup didn't improve my score both local and LB. Does it works for you?", "votes": null }, { "id": "1157828", "postDate": "01/18/2021 07:22:57", "content": "

thanks for letting me know @wantsu

\n

i am still trying other augmentations like cutmix
\ni will let you know when i'll try mixup,thank you

\n

note : if you are training pure mixup and not \"mixup without hesitation\" implementation then you need to go for long training for this algo to converge(pure mixup is slow)

", "rawMarkdown": "thanks for letting me know @wantsu \n\ni am still trying other augmentations like cutmix\ni will let you know when i'll try mixup,thank you\n\nnote : if you are training pure mixup and not \"mixup without hesitation\" implementation then you need to go for long training for this algo to converge(pure mixup is slow)", "votes": null }, { "id": "1157902", "postDate": "01/18/2021 08:37:40", "content": "

Thanks for your sharing of this paper @Mobassir. I used the pure mixup to train my model, and will try this improved version.

", "rawMarkdown": "Thanks for your sharing of this paper @Mobassir. I used the pure mixup to train my model, and will try this improved version.", "votes": null }, { "id": "1157980", "postDate": "01/18/2021 09:59:54", "content": "

I've tried combination of CutMix[P(0.5x2/3)] and MixUp[P(0.5x2/3)] and both of them helped me improve my Cv as well as Lb.

", "rawMarkdown": "I've tried combination of *CutMix[P(0.5x2/3)] and MixUp[P(0.5x2/3)]* and both of them helped me improve my Cv as well as Lb.", "votes": null }, { "id": "1157981", "postDate": "01/18/2021 10:00:51", "content": "

dear @thakurudit
\nthanks for letting us know,best of luck for this competition

", "rawMarkdown": "dear @thakurudit \nthanks for letting us know,best of luck for this competition", "votes": null } ], "comments": [ { "id": 1157780, "author_name": "wantsu", "author_url": "", "post_date": "01/18/2021 06:22:38", "content": "

Hello, in my experiment, Mixup didn't improve my score both local and LB. Does it works for you?

", "votes": null, "replies": [ { "id": 1157828, "author_name": "mobassir", "author_url": "", "post_date": "01/18/2021 07:22:57", "content": "

thanks for letting me know @wantsu

\n

i am still trying other augmentations like cutmix
\ni will let you know when i'll try mixup,thank you

\n

note : if you are training pure mixup and not \"mixup without hesitation\" implementation then you need to go for long training for this algo to converge(pure mixup is slow)

", "votes": null, "replies": [] }, { "id": 1157902, "author_name": "wantsu", "author_url": "", "post_date": "01/18/2021 08:37:40", "content": "

Thanks for your sharing of this paper @Mobassir. I used the pure mixup to train my model, and will try this improved version.

", "votes": null, "replies": [] } ] }, { "id": 1157980, "author_name": "thakurudit", "author_url": "", "post_date": "01/18/2021 09:59:54", "content": "

I've tried combination of CutMix[P(0.5x2/3)] and MixUp[P(0.5x2/3)] and both of them helped me improve my Cv as well as Lb.

", "votes": null, "replies": [ { "id": 1157981, "author_name": "mobassir", "author_url": "", "post_date": "01/18/2021 10:00:51", "content": "

dear @thakurudit
\nthanks for letting us know,best of luck for this competition

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1156676": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2034058%2F9640da7324cea7d6953221bd443259c8%2Fmix.jpg?generation=1610878944120116&alt=media)\n\n\n\n\n# paper : https://arxiv.org/abs/2101.04342\n\n# code : https://github.com/yuhao318/mwh", "1157780": "Hello, in my experiment, Mixup didn't improve my score both local and LB. Does it works for you?", "1157828": "thanks for letting me know @wantsu \n\ni am still trying other augmentations like cutmix\ni will let you know when i'll try mixup,thank you\n\nnote : if you are training pure mixup and not \"mixup without hesitation\" implementation then you need to go for long training for this algo to converge(pure mixup is slow)", "1157902": "Thanks for your sharing of this paper @Mobassir. I used the pure mixup to train my model, and will try this improved version.", "1157980": "I've tried combination of *CutMix[P(0.5x2/3)] and MixUp[P(0.5x2/3)]* and both of them helped me improve my Cv as well as Lb.", "1157981": "dear @thakurudit \nthanks for letting us know,best of luck for this competition" }, "source": "meta" }