{ "id": 340872, "title": "Using HuBMAP Kidney 2021's data as external data", "url": "/competitions/hubmap-organ-segmentation/discussion/340872", "author_name": "", "post_date": "2022-07-31T10:53:34.868048Z", "votes": 10, "comment_count": 5, "views": 0, "content": "

HubMAP - Hacking the Kidney 2021 data as external data

\n

Here is a link to the HuBMAP 2021 - Hacking the Kidney's 1024x1024 PNG dataset for external data purposes:
\nhttps://www.kaggle.com/datasets/kfk42kfk/kidney2trn-1

\n

About this dataset

\n

This dataset is created using @cdeotte's \"Balanced Stochastic Data Loader\" from the previous HuBMAP competition.
\nPNG images are cropped from big tiff files and converted to 1024x1024 tiles.
\nAlso by using the \"Balanced Stochastic Data Loader\" every tile has a segmentation mask.

\n
\n

Lastly, the data loader guarantees that each crop of 1024x1024 contains at least 1 segmentation label. (If not, randomly select again).
\n from: https://www.kaggle.com/competitions/hubmap-kidney-segmentation/discussion/238308

\n
\n

PS: Both masks and images are not normalized. They both contain [0-255] pixel values. No augmentations used.

", "messages": [ { "id": "1878285", "postDate": "07/31/2022 10:53:34", "content": "

HubMAP - Hacking the Kidney 2021 data as external data

\n

Here is a link to the HuBMAP 2021 - Hacking the Kidney's 1024x1024 PNG dataset for external data purposes:
\nhttps://www.kaggle.com/datasets/kfk42kfk/kidney2trn-1

\n

About this dataset

\n

This dataset is created using @cdeotte's \"Balanced Stochastic Data Loader\" from the previous HuBMAP competition.
\nPNG images are cropped from big tiff files and converted to 1024x1024 tiles.
\nAlso by using the \"Balanced Stochastic Data Loader\" every tile has a segmentation mask.

\n
\n

Lastly, the data loader guarantees that each crop of 1024x1024 contains at least 1 segmentation label. (If not, randomly select again).
\n from: https://www.kaggle.com/competitions/hubmap-kidney-segmentation/discussion/238308

\n
\n

PS: Both masks and images are not normalized. They both contain [0-255] pixel values. No augmentations used.

", "rawMarkdown": "## HubMAP - Hacking the Kidney 2021 data as external data\nHere is a link to the HuBMAP 2021 - Hacking the Kidney's 1024x1024 PNG dataset for external data purposes:\nhttps://www.kaggle.com/datasets/kfk42kfk/kidney2trn-1\n\n## About this dataset\nThis dataset is created using @cdeotte's \"Balanced Stochastic Data Loader\" from the previous HuBMAP competition.\nPNG images are cropped from big tiff files and converted to 1024x1024 tiles.\nAlso by using the \"Balanced Stochastic Data Loader\" every tile has a segmentation mask. \n> Lastly, the data loader guarantees that each crop of 1024x1024 contains at least 1 segmentation label. (If not, randomly select again). \nfrom: https://www.kaggle.com/competitions/hubmap-kidney-segmentation/discussion/238308\n\n\nPS: Both masks and images are not normalized. They both contain [0-255] pixel values. No augmentations used.", "votes": null }, { "id": "1878956", "postDate": "07/31/2022 18:52:24", "content": "

Do you know the pixel scale of those images? Like 0.65?

", "rawMarkdown": "Do you know the pixel scale of those images? Like 0.65?", "votes": null }, { "id": "1880580", "postDate": "08/01/2022 20:28:51", "content": "

A concern

\n

Including HuBMAP 2021's data to actual training data and training with the mixed data might cause (or would cause?) unbalanced dataset issues. But since we are predicting the same thing, maybe it wouldn't be a problem(?).

", "rawMarkdown": "### **A concern**\nIncluding HuBMAP 2021's data to actual training data and training with the mixed data might cause (or would cause?) unbalanced dataset issues. But since we are predicting the same thing, maybe it wouldn't be a problem(?).", "votes": null }, { "id": "1880668", "postDate": "08/01/2022 23:36:04", "content": "

\" might\" cause

\n

just verify by experiments

", "rawMarkdown": "\" might\" cause \n\njust verify by experiments", "votes": null }, { "id": "1887002", "postDate": "08/06/2022 11:03:22", "content": "

using kidney 2021 dataset + zenodo dataset (both of them are labeled), didn't help improving LB even though it improved CV a lot. It might because of both of the added datasets are kidney datasets, and this is probably upsetting the balance of organs.

", "rawMarkdown": "using kidney 2021 dataset + zenodo dataset (both of them are labeled), didn't help improving LB even though it improved CV a lot. It might because of both of the added datasets are kidney datasets, and this is probably upsetting the balance of organs.", "votes": null }, { "id": "1889428", "postDate": "08/08/2022 07:22:31", "content": "

what is the meaning of pixel scale?

", "rawMarkdown": "what is the meaning of pixel scale?", "votes": null } ], "comments": [ { "id": 1878956, "author_name": "sakvaua", "author_url": "", "post_date": "07/31/2022 18:52:24", "content": "

Do you know the pixel scale of those images? Like 0.65?

", "votes": null, "replies": [ { "id": 1889428, "author_name": "tanxxx", "author_url": "", "post_date": "08/08/2022 07:22:31", "content": "

what is the meaning of pixel scale?

", "votes": null, "replies": [] } ] }, { "id": 1880580, "author_name": "kfk42kfk", "author_url": "", "post_date": "08/01/2022 20:28:51", "content": "

A concern

\n

Including HuBMAP 2021's data to actual training data and training with the mixed data might cause (or would cause?) unbalanced dataset issues. But since we are predicting the same thing, maybe it wouldn't be a problem(?).

", "votes": null, "replies": [ { "id": 1880668, "author_name": "hengck23", "author_url": "", "post_date": "08/01/2022 23:36:04", "content": "

\" might\" cause

\n

just verify by experiments

", "votes": null, "replies": [] }, { "id": 1887002, "author_name": "kfk42kfk", "author_url": "", "post_date": "08/06/2022 11:03:22", "content": "

using kidney 2021 dataset + zenodo dataset (both of them are labeled), didn't help improving LB even though it improved CV a lot. It might because of both of the added datasets are kidney datasets, and this is probably upsetting the balance of organs.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1878285": "## HubMAP - Hacking the Kidney 2021 data as external data\nHere is a link to the HuBMAP 2021 - Hacking the Kidney's 1024x1024 PNG dataset for external data purposes:\nhttps://www.kaggle.com/datasets/kfk42kfk/kidney2trn-1\n\n## About this dataset\nThis dataset is created using @cdeotte's \"Balanced Stochastic Data Loader\" from the previous HuBMAP competition.\nPNG images are cropped from big tiff files and converted to 1024x1024 tiles.\nAlso by using the \"Balanced Stochastic Data Loader\" every tile has a segmentation mask. \n> Lastly, the data loader guarantees that each crop of 1024x1024 contains at least 1 segmentation label. (If not, randomly select again). \nfrom: https://www.kaggle.com/competitions/hubmap-kidney-segmentation/discussion/238308\n\n\nPS: Both masks and images are not normalized. They both contain [0-255] pixel values. No augmentations used.", "1878956": "Do you know the pixel scale of those images? Like 0.65?", "1880580": "### **A concern**\nIncluding HuBMAP 2021's data to actual training data and training with the mixed data might cause (or would cause?) unbalanced dataset issues. But since we are predicting the same thing, maybe it wouldn't be a problem(?).", "1880668": "\" might\" cause \n\njust verify by experiments", "1887002": "using kidney 2021 dataset + zenodo dataset (both of them are labeled), didn't help improving LB even though it improved CV a lot. It might because of both of the added datasets are kidney datasets, and this is probably upsetting the balance of organs.", "1889428": "what is the meaning of pixel scale?" }, "source": "meta" }