![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F286455%2Fbcf95909ea81b9b6daeb847ed47f445c%2FScreenshot%202024-02-17%20at%2013.09.01.png?generation=1708219213032529&alt=media\")
As I understand from the SpecAugment paper, Time and Frequency masking is a useful augmentation for training on Spectrograms.
\nToday, we released Albumentations 1.4.0, where we Added XYMasking transform, a generalization of TimeMasking and FrequencyMasking from torchaudio.
\nHere is Kaggle Kernel with examples
\n
We hope it will help in this competition.
\nP.S. I have one request - if you want to say \"thank you\", do not hesitate to add a \"star\" on: the repository page, if not - it is ok, just use the library and try to get to the top of the leaderboard ;)
", "messages": [ { "id": "2656734", "postDate": "02/18/2024 01:22:50", "content": "As I understand from the SpecAugment paper, Time and Frequency masking is a useful augmentation for training on Spectrograms.
\nToday, we released Albumentations 1.4.0, where we Added XYMasking transform, a generalization of TimeMasking and FrequencyMasking from torchaudio.
\nHere is Kaggle Kernel with examples
\n
We hope it will help in this competition.
\nP.S. I have one request - if you want to say \"thank you\", do not hesitate to add a \"star\" on: the repository page, if not - it is ok, just use the library and try to get to the top of the leaderboard ;)
", "rawMarkdown": "As I understand from the SpecAugment paper, Time and Frequency masking is a useful augmentation for training on Spectrograms.\n\nToday, we [released Albumentations 1.4.0](https://twitter.com/albumentations), where we Added XYMasking transform, a generalization of TimeMasking and FrequencyMasking from torchaudio. \n\nHere is [Kaggle Kernel with examples](https://www.kaggle.com/code/iglovikov/xymasking-aug)\n\n\n\nWe hope it will help in this competition.\n\nP.S. I have one request - if you want to say \"thank you\", do not hesitate to add a \"star\" on: [the repository page](https://github.com/albumentations-team/albumentations/), if not - it is ok, just use the library and try to get to the top of the leaderboard ;)", "votes": null }, { "id": "2656747", "postDate": "02/18/2024 01:46:20", "content": "good job! I am considering improving my data augmentation strategy. I will report the effectiveness of the XYMasking transform later.
", "rawMarkdown": "good job! I am considering improving my data augmentation strategy. I will report the effectiveness of the XYMasking transform later.", "votes": null }, { "id": "2656762", "postDate": "02/18/2024 02:08:29", "content": "This is pretty cool! Looking forward to testing it out! Makes my augmentation pipeline so much cleaner.
", "rawMarkdown": "This is pretty cool! Looking forward to testing it out! Makes my augmentation pipeline so much cleaner.", "votes": null }, { "id": "2656763", "postDate": "02/18/2024 02:08:44", "content": "Thanks! If it will work for you, maybe I will try to trainsomething here as well
", "rawMarkdown": "Thanks! If it will work for you, maybe I will try to trainsomething here as well", "votes": null }, { "id": "2656765", "postDate": "02/18/2024 02:10:08", "content": "You are welcome.
\nI have a request - if next time you have in mind a transform that is not implemented in Albumentations, just let me know, and we will figure something out.
", "rawMarkdown": "You are welcome.\n\nI have a request - if next time you have in mind a transform that is not implemented in Albumentations, just let me know, and we will figure something out.", "votes": null }, { "id": "2656778", "postDate": "02/18/2024 02:31:54", "content": "I'm testing it out now :)
\nIs mixup something that could be added to Albumentations?
", "rawMarkdown": "I'm testing it out now :)\n\nIs mixup something that could be added to Albumentations?", "votes": null }, { "id": "2656816", "postDate": "02/18/2024 03:38:50", "content": "I think so, I will try to add it next week.
", "rawMarkdown": "I think so, I will try to add it next week.", "votes": null }, { "id": "2656821", "postDate": "02/18/2024 03:46:35", "content": "That would be awesome :). FYI, the time and frequency masking is working great for me. I need to see how it impacts CV and what the right values are, but it dropped right into my pipeline very smoothly.
", "rawMarkdown": "That would be awesome :). FYI, the time and frequency masking is working great for me. I need to see how it impacts CV and what the right values are, but it dropped right into my pipeline very smoothly.", "votes": null }, { "id": "2665905", "postDate": "02/24/2024 00:43:29", "content": "I have a question about updating the package. As far as I understand, the competition prohibits Internet access. So I can't update the package while submitting. Is there any way to submit notebook environment with updated albumentations package?
", "rawMarkdown": "I have a question about updating the package. As far as I understand, the competition prohibits Internet access. So I can't update the package while submitting. Is there any way to submit notebook environment with updated albumentations package?", "votes": null }, { "id": "2670209", "postDate": "02/26/2024 17:50:26", "content": "You would have to upload the package as a wheel and attach the dataset to your notebook. Then you can install it from there when you run your offline notebook.
", "rawMarkdown": "You would have to upload the package as a wheel and attach the dataset to your notebook. Then you can install it from there when you run your offline notebook.", "votes": null }, { "id": "2688346", "postDate": "03/09/2024 07:09:23", "content": "hi, i have two questions:
\nHow are both EEG data and spectrogram incorporated into model training processes?
Is it necessary to implement any filtering, denoising, or normalization techniques on the provided readings?
good job! I am considering improving my data augmentation strategy. I will report the effectiveness of the XYMasking transform later.
", "votes": null, "replies": [ { "id": 2656763, "author_name": "iglovikov", "author_url": "", "post_date": "02/18/2024 02:08:44", "content": "Thanks! If it will work for you, maybe I will try to trainsomething here as well
", "votes": null, "replies": [] } ] }, { "id": 2656762, "author_name": "chemdatafarmer", "author_url": "", "post_date": "02/18/2024 02:08:29", "content": "This is pretty cool! Looking forward to testing it out! Makes my augmentation pipeline so much cleaner.
", "votes": null, "replies": [ { "id": 2656765, "author_name": "iglovikov", "author_url": "", "post_date": "02/18/2024 02:10:08", "content": "You are welcome.
\nI have a request - if next time you have in mind a transform that is not implemented in Albumentations, just let me know, and we will figure something out.
", "votes": null, "replies": [ { "id": 2656778, "author_name": "chemdatafarmer", "author_url": "", "post_date": "02/18/2024 02:31:54", "content": "I'm testing it out now :)
\nIs mixup something that could be added to Albumentations?
", "votes": null, "replies": [ { "id": 2656816, "author_name": "iglovikov", "author_url": "", "post_date": "02/18/2024 03:38:50", "content": "I think so, I will try to add it next week.
", "votes": null, "replies": [ { "id": 2656821, "author_name": "chemdatafarmer", "author_url": "", "post_date": "02/18/2024 03:46:35", "content": "That would be awesome :). FYI, the time and frequency masking is working great for me. I need to see how it impacts CV and what the right values are, but it dropped right into my pipeline very smoothly.
", "votes": null, "replies": [] } ] } ] } ] } ] }, { "id": 2665905, "author_name": "yerassylbolatov", "author_url": "", "post_date": "02/24/2024 00:43:29", "content": "I have a question about updating the package. As far as I understand, the competition prohibits Internet access. So I can't update the package while submitting. Is there any way to submit notebook environment with updated albumentations package?
", "votes": null, "replies": [ { "id": 2670209, "author_name": "janbrederecke", "author_url": "", "post_date": "02/26/2024 17:50:26", "content": "You would have to upload the package as a wheel and attach the dataset to your notebook. Then you can install it from there when you run your offline notebook.
", "votes": null, "replies": [] } ] }, { "id": 2688346, "author_name": "farwa99", "author_url": "", "post_date": "03/09/2024 07:09:23", "content": "hi, i have two questions:
\nHow are both EEG data and spectrogram incorporated into model training processes?
Is it necessary to implement any filtering, denoising, or normalization techniques on the provided readings?