I run into memory limits trying to load in the full brain (or half of a brain) for even a single channel, with a batch size of 1, and I believe this is hamstringing my model's efforts to learn. If there were a good way to identify the 10 or so slices in each mode that contain the most tumor, it would make it much more efficient to train a model.
", "messages": [ { "id": "1526191", "postDate": "09/27/2021 23:51:28", "content": "I run into memory limits trying to load in the full brain (or half of a brain) for even a single channel, with a batch size of 1, and I believe this is hamstringing my model's efforts to learn. If there were a good way to identify the 10 or so slices in each mode that contain the most tumor, it would make it much more efficient to train a model.
", "rawMarkdown": "I run into memory limits trying to load in the full brain (or half of a brain) for even a single channel, with a batch size of 1, and I believe this is hamstringing my model's efforts to learn. If there were a good way to identify the 10 or so slices in each mode that contain the most tumor, it would make it much more efficient to train a model.", "votes": null }, { "id": "1526367", "postDate": "09/28/2021 04:15:34", "content": "Check this notebook https://www.kaggle.com/josecarmona/btrc-eda-final/notebook.
", "rawMarkdown": "Check this notebook https://www.kaggle.com/josecarmona/btrc-eda-final/notebook.", "votes": null }, { "id": "1526825", "postDate": "09/28/2021 11:10:21", "content": "HI @maxbaugh first you should have to do segmentation, I have tried in my notebook https://www.kaggle.com/victorfernandezalbor/brats-20-win-nnunet-segment-with-brats-21-rsna with brats data, but of course you can choose you favorite way, best of luck!
", "rawMarkdown": "HI @maxbaugh first you should have to do segmentation, I have tried in my notebook https://www.kaggle.com/victorfernandezalbor/brats-20-win-nnunet-segment-with-brats-21-rsna with brats data, but of course you can choose you favorite way, best of luck!", "votes": null }, { "id": "1530158", "postDate": "10/01/2021 01:32:58", "content": "I made a notebook for Object Detection of tumors in all three planes of the T1wCE series using YOLOv5. It's pretty accurate if you exclude empty images and some of the peripheral slices. It might help.
\nhttps://www.kaggle.com/davidbroberts/brain-tumor-object-detection
", "rawMarkdown": "I made a notebook for Object Detection of tumors in all three planes of the T1wCE series using YOLOv5. It's pretty accurate if you exclude empty images and some of the peripheral slices. It might help.\n\nhttps://www.kaggle.com/davidbroberts/brain-tumor-object-detection", "votes": null }, { "id": "1530981", "postDate": "10/01/2021 14:14:24", "content": "How is it going? I'm not very good with mask rcnn?
", "rawMarkdown": "How is it going? I'm not very good with mask rcnn?", "votes": null } ], "comments": [ { "id": 1526367, "author_name": "stelis", "author_url": "", "post_date": "09/28/2021 04:15:34", "content": "Check this notebook https://www.kaggle.com/josecarmona/btrc-eda-final/notebook.
", "votes": null, "replies": [] }, { "id": 1526825, "author_name": "victorfernandezalbor", "author_url": "", "post_date": "09/28/2021 11:10:21", "content": "HI @maxbaugh first you should have to do segmentation, I have tried in my notebook https://www.kaggle.com/victorfernandezalbor/brats-20-win-nnunet-segment-with-brats-21-rsna with brats data, but of course you can choose you favorite way, best of luck!
", "votes": null, "replies": [] }, { "id": 1530158, "author_name": "davidbroberts", "author_url": "", "post_date": "10/01/2021 01:32:58", "content": "I made a notebook for Object Detection of tumors in all three planes of the T1wCE series using YOLOv5. It's pretty accurate if you exclude empty images and some of the peripheral slices. It might help.
\nhttps://www.kaggle.com/davidbroberts/brain-tumor-object-detection
", "votes": null, "replies": [ { "id": 1530981, "author_name": "zaakciiru", "author_url": "", "post_date": "10/01/2021 14:14:24", "content": "How is it going? I'm not very good with mask rcnn?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1526191": "I run into memory limits trying to load in the full brain (or half of a brain) for even a single channel, with a batch size of 1, and I believe this is hamstringing my model's efforts to learn. If there were a good way to identify the 10 or so slices in each mode that contain the most tumor, it would make it much more efficient to train a model.", "1526367": "Check this notebook https://www.kaggle.com/josecarmona/btrc-eda-final/notebook.", "1526825": "HI @maxbaugh first you should have to do segmentation, I have tried in my notebook https://www.kaggle.com/victorfernandezalbor/brats-20-win-nnunet-segment-with-brats-21-rsna with brats data, but of course you can choose you favorite way, best of luck!", "1530158": "I made a notebook for Object Detection of tumors in all three planes of the T1wCE series using YOLOv5. It's pretty accurate if you exclude empty images and some of the peripheral slices. It might help.\n\nhttps://www.kaggle.com/davidbroberts/brain-tumor-object-detection", "1530981": "How is it going? I'm not very good with mask rcnn?" }, "source": "meta" }