Hello Kagglers! 👋
\nI've got a 0.65 LB score using a single fold model with 12 channels, and I'm excited to share my approach with you! 🎉
\nCheck out my RSNA24 Starter Notebook using TensorFlow to dive in.
\nrow_id using 12 image channels: You can also train helper models to predict:
\nThese models are already trained and included with an impressive accuracy of over 95%! 🚀
\nThe main constraint is GPU and memory, so if you've got enough of those, this should work out great for you.
\nI'm also open to teaming up! If you're interested, feel free to DM me. 😊
\nIf you find my work helpful, don't forget to give it an upvote. Thanks for checking it out!
\nHappy modeling! 🎯
", "messages": [ { "id": "2983918", "postDate": "09/09/2024 08:20:37", "content": "Hello Kagglers! 👋
\nI've got a 0.65 LB score using a single fold model with 12 channels, and I'm excited to share my approach with you! 🎉
\nCheck out my RSNA24 Starter Notebook using TensorFlow to dive in.
\nrow_id using 12 image channels: You can also train helper models to predict:
\nThese models are already trained and included with an impressive accuracy of over 95%! 🚀
\nThe main constraint is GPU and memory, so if you've got enough of those, this should work out great for you.
\nI'm also open to teaming up! If you're interested, feel free to DM me. 😊
\nIf you find my work helpful, don't forget to give it an upvote. Thanks for checking it out!
\nHappy modeling! 🎯
", "rawMarkdown": "Hello Kagglers! 👋\n\nI've got a 0.65 LB score using a single fold model with 12 channels, and I'm excited to share my approach with you! 🎉\n\nCheck out my [RSNA24 Starter Notebook using TensorFlow](https://www.kaggle.com/code/nartaa/rsna24-starter-tf) to dive in.\n\n### Model Overview:\n- The base model is designed to train and predict per `row_id` using 12 image channels: \n - **4 Axial**\n - **4 Sagittal**\n - **4 Sagittal STIR**\n- The code is flexible, so you can increase the number of channels—just make sure it's divisible by 3.\n\n### Helper Models:\nYou can also train helper models to predict:\n- **Axial Level**\n- **Sagittal Side**\n\nThese models are already trained and included with an impressive accuracy of over 95%! 🚀\n\n### Limitations:\nThe main constraint is GPU and memory, so if you've got enough of those, this should work out great for you. \n\nI'm also open to teaming up! If you're interested, feel free to DM me. 😊\n\nIf you find my work helpful, don't forget to give it an upvote. Thanks for checking it out!\n\nHappy modeling! 🎯", "votes": null }, { "id": "2990717", "postDate": "09/16/2024 16:23:27", "content": "thanks man, but is there an alternative to describe row_id(Axial Level ,Sagittal Side…) for submission of a model that only predicts condition
", "rawMarkdown": "thanks man, but is there an alternative to describe row_id(Axial Level ,Sagittal Side...) for submission of a model that only predicts condition", "votes": null }, { "id": "2990728", "postDate": "09/16/2024 16:35:11", "content": "you're welcome.
\nI am not sure I follow, could you elaborate.
i mean ,could there be a way to name the predicted image(row_id) as reqiured in the submission file like \"study_id_[condition]_[level]\" suppose one has designed a model that predicts only 3 classes [Normal/Mild, Moderate, or Severe] because i see your solution simplifies this problem but at a computational cost (by predicting Axial Level and as well as Sagittal Side) that way it will then be easy to just join the sudy_id with the predicted condition and level to come up with the expected row_id e.g 44036939_left_neural_foraminal_narrowing_l1_l2 or the full information about the row_id could be available in the .dcm meta information ?
", "rawMarkdown": "i mean ,could there be a way to name the predicted image(row_id) as reqiured in the submission file like \"study_id_[condition]_[level]\" suppose one has designed a model that predicts only 3 classes [Normal/Mild, Moderate, or Severe] because i see your solution simplifies this problem but at a computational cost (by predicting Axial Level and as well as Sagittal Side) that way it will then be easy to just join the sudy_id with the predicted condition and level to come up with the expected row_id e.g 44036939_left_neural_foraminal_narrowing_l1_l2 or the full information about the row_id could be available in the .dcm meta information ?", "votes": null }, { "id": "2990881", "postDate": "09/16/2024 18:46:59", "content": "I don't see how this would work, but if you come up with a code feel free to share it.
", "rawMarkdown": "I don't see how this would work, but if you come up with a code feel free to share it.", "votes": null } ], "comments": [ { "id": 2990717, "author_name": "keniasnyandoro", "author_url": "", "post_date": "09/16/2024 16:23:27", "content": "thanks man, but is there an alternative to describe row_id(Axial Level ,Sagittal Side…) for submission of a model that only predicts condition
", "votes": null, "replies": [ { "id": 2990728, "author_name": "nartaa", "author_url": "", "post_date": "09/16/2024 16:35:11", "content": "you're welcome.
\nI am not sure I follow, could you elaborate.
i mean ,could there be a way to name the predicted image(row_id) as reqiured in the submission file like \"study_id_[condition]_[level]\" suppose one has designed a model that predicts only 3 classes [Normal/Mild, Moderate, or Severe] because i see your solution simplifies this problem but at a computational cost (by predicting Axial Level and as well as Sagittal Side) that way it will then be easy to just join the sudy_id with the predicted condition and level to come up with the expected row_id e.g 44036939_left_neural_foraminal_narrowing_l1_l2 or the full information about the row_id could be available in the .dcm meta information ?
", "votes": null, "replies": [ { "id": 2990881, "author_name": "nartaa", "author_url": "", "post_date": "09/16/2024 18:46:59", "content": "I don't see how this would work, but if you come up with a code feel free to share it.
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "2983918": "Hello Kagglers! 👋\n\nI've got a 0.65 LB score using a single fold model with 12 channels, and I'm excited to share my approach with you! 🎉\n\nCheck out my [RSNA24 Starter Notebook using TensorFlow](https://www.kaggle.com/code/nartaa/rsna24-starter-tf) to dive in.\n\n### Model Overview:\n- The base model is designed to train and predict per `row_id` using 12 image channels: \n - **4 Axial**\n - **4 Sagittal**\n - **4 Sagittal STIR**\n- The code is flexible, so you can increase the number of channels—just make sure it's divisible by 3.\n\n### Helper Models:\nYou can also train helper models to predict:\n- **Axial Level**\n- **Sagittal Side**\n\nThese models are already trained and included with an impressive accuracy of over 95%! 🚀\n\n### Limitations:\nThe main constraint is GPU and memory, so if you've got enough of those, this should work out great for you. \n\nI'm also open to teaming up! If you're interested, feel free to DM me. 😊\n\nIf you find my work helpful, don't forget to give it an upvote. Thanks for checking it out!\n\nHappy modeling! 🎯", "2990717": "thanks man, but is there an alternative to describe row_id(Axial Level ,Sagittal Side...) for submission of a model that only predicts condition", "2990728": "you're welcome.\nI am not sure I follow, could you elaborate.", "2990769": "i mean ,could there be a way to name the predicted image(row_id) as reqiured in the submission file like \"study_id_[condition]_[level]\" suppose one has designed a model that predicts only 3 classes [Normal/Mild, Moderate, or Severe] because i see your solution simplifies this problem but at a computational cost (by predicting Axial Level and as well as Sagittal Side) that way it will then be easy to just join the sudy_id with the predicted condition and level to come up with the expected row_id e.g 44036939_left_neural_foraminal_narrowing_l1_l2 or the full information about the row_id could be available in the .dcm meta information ?", "2990881": "I don't see how this would work, but if you come up with a code feel free to share it." }, "source": "meta" }