{ "id": 332921, "title": "HuBMAP + HPA Starter kit", "url": "/competitions/hubmap-organ-segmentation/discussion/332921", "author_name": "", "post_date": "2022-06-23T23:31:44.645699Z", "votes": 11, "comment_count": 2, "views": 0, "content": "

Hi everyone,

\n

Just to let you know, I created a series of notebooks & datasets that might be of interest as a starting point.

\n
\n

Credits
\nAll notebooks are made based on the incredible series of notebooks by iafoss.
\nAll credit goes to him!

\n
\n
Notebooks
\n

Modeling - FastAI

\n
[Training] - FastAI Baseline
\n
[Inference] - FastAI Baseline
\n
\n

Dataset Creation

\n

All models use a preprocessed dataset I made using the following notebook:

\n
https://www.kaggle.com/code/thedevastator/converting-to-256x256
\n
\n

EDA

\n

And also here is a quick EDA notebook just for getting started:

\n
Quick EDA
\n
\n
Datasets
\n

Also, I already created some scaled versions of the dataset (using the notebook above) so you can play around with:

\n
256 x 256
\n
512 x 512
\n
128 x 128
\n
\n
\n

Please note: Fixed all prediction issues. Have fun!

\n
", "messages": [ { "id": "1831075", "postDate": "06/23/2022 23:31:44", "content": "

Hi everyone,

\n

Just to let you know, I created a series of notebooks & datasets that might be of interest as a starting point.

\n
\n

Credits
\nAll notebooks are made based on the incredible series of notebooks by iafoss.
\nAll credit goes to him!

\n
\n
Notebooks
\n

Modeling - FastAI

\n
[Training] - FastAI Baseline
\n
[Inference] - FastAI Baseline
\n
\n

Dataset Creation

\n

All models use a preprocessed dataset I made using the following notebook:

\n
https://www.kaggle.com/code/thedevastator/converting-to-256x256
\n
\n

EDA

\n

And also here is a quick EDA notebook just for getting started:

\n
Quick EDA
\n
\n
Datasets
\n

Also, I already created some scaled versions of the dataset (using the notebook above) so you can play around with:

\n
256 x 256
\n
512 x 512
\n
128 x 128
\n
\n
\n

Please note: Fixed all prediction issues. Have fun!

\n
", "rawMarkdown": "Hi everyone, \n\nJust to let you know, I created a series of notebooks & datasets that might be of interest as a starting point. \n\n_____\n**Credits**\nAll notebooks are made based on the [incredible](https://www.kaggle.com/code/iafoss/256x256-images) [series](https://www.kaggle.com/code/iafoss/hubmap-pytorch-fast-ai-starter-sub) [of](https://www.kaggle.com/code/iafoss/hubmap-pytorch-fast-ai-starter) notebooks by [iafoss](https://www.kaggle.com/iafoss). \nAll credit goes to him! \n_____\n\n\n##### Notebooks\n\n**Modeling - FastAI**\n##### [[Training] - FastAI Baseline](https://www.kaggle.com/code/thedevastator/training-fastai-baseline)\n##### [[Inference] - FastAI Baseline](https://www.kaggle.com/code/thedevastator/inference-fastai-baseline)\n_____\n**Dataset Creation**\n\nAll models use a preprocessed dataset I made using the following notebook: \n\n##### [https://www.kaggle.com/code/thedevastator/converting-to-256x256](https://www.kaggle.com/code/thedevastator/converting-to-256x256)\n_____\n\n**EDA**\n\nAnd also here is a quick EDA notebook just for getting started:\n##### [Quick EDA](https://www.kaggle.com/code/thedevastator/hubmap-quick-eda)\n_____\n##### Datasets\n\nAlso, I already created some scaled versions of the dataset (using the notebook above) so you can play around with:\n\n##### [256 x 256](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-256x256)\n##### [512 x 512](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-512x512)\n##### [128 x 128](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-128x128)\n\n_____\n\n> **Please note:** Fixed all prediction issues. Have fun!", "votes": null }, { "id": "1831227", "postDate": "06/24/2022 04:00:45", "content": "

Curious about the 512x512 dataset you have created, I see that you have cropped the image and its mask into 8 regions why not just resize them as you will be loosing global semantic info if you train on them. Also the test image will be very different from these images. Any particular reason for doing so?

", "rawMarkdown": "Curious about the 512x512 dataset you have created, I see that you have cropped the image and its mask into 8 regions why not just resize them as you will be loosing global semantic info if you train on them. Also the test image will be very different from these images. Any particular reason for doing so?", "votes": null }, { "id": "1831949", "postDate": "06/24/2022 14:34:42", "content": "

@thedevastator Great work
\nI’ve checked your inference notebook and I know why you got an error

\n

It’s the submission file header it should be id, rle not “predicted”

", "rawMarkdown": "thedevastator Great work\nI’ve checked your inference notebook and I know why you got an error\n\nIt’s the submission file header it should be id, rle not “predicted”", "votes": null } ], "comments": [ { "id": 1831227, "author_name": "nishantbhansali", "author_url": "", "post_date": "06/24/2022 04:00:45", "content": "

Curious about the 512x512 dataset you have created, I see that you have cropped the image and its mask into 8 regions why not just resize them as you will be loosing global semantic info if you train on them. Also the test image will be very different from these images. Any particular reason for doing so?

", "votes": null, "replies": [] }, { "id": 1831949, "author_name": "asalhi", "author_url": "", "post_date": "06/24/2022 14:34:42", "content": "

@thedevastator Great work
\nI’ve checked your inference notebook and I know why you got an error

\n

It’s the submission file header it should be id, rle not “predicted”

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1831075": "Hi everyone, \n\nJust to let you know, I created a series of notebooks & datasets that might be of interest as a starting point. \n\n_____\n**Credits**\nAll notebooks are made based on the [incredible](https://www.kaggle.com/code/iafoss/256x256-images) [series](https://www.kaggle.com/code/iafoss/hubmap-pytorch-fast-ai-starter-sub) [of](https://www.kaggle.com/code/iafoss/hubmap-pytorch-fast-ai-starter) notebooks by [iafoss](https://www.kaggle.com/iafoss). \nAll credit goes to him! \n_____\n\n\n##### Notebooks\n\n**Modeling - FastAI**\n##### [[Training] - FastAI Baseline](https://www.kaggle.com/code/thedevastator/training-fastai-baseline)\n##### [[Inference] - FastAI Baseline](https://www.kaggle.com/code/thedevastator/inference-fastai-baseline)\n_____\n**Dataset Creation**\n\nAll models use a preprocessed dataset I made using the following notebook: \n\n##### [https://www.kaggle.com/code/thedevastator/converting-to-256x256](https://www.kaggle.com/code/thedevastator/converting-to-256x256)\n_____\n\n**EDA**\n\nAnd also here is a quick EDA notebook just for getting started:\n##### [Quick EDA](https://www.kaggle.com/code/thedevastator/hubmap-quick-eda)\n_____\n##### Datasets\n\nAlso, I already created some scaled versions of the dataset (using the notebook above) so you can play around with:\n\n##### [256 x 256](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-256x256)\n##### [512 x 512](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-512x512)\n##### [128 x 128](https://www.kaggle.com/datasets/thedevastator/hubmap-2022-128x128)\n\n_____\n\n> **Please note:** Fixed all prediction issues. Have fun!", "1831227": "Curious about the 512x512 dataset you have created, I see that you have cropped the image and its mask into 8 regions why not just resize them as you will be loosing global semantic info if you train on them. Also the test image will be very different from these images. Any particular reason for doing so?", "1831949": "thedevastator Great work\nI’ve checked your inference notebook and I know why you got an error\n\nIt’s the submission file header it should be id, rle not “predicted”" }, "source": "meta" }