{ "id": 232357, "title": "Fast Code to Convert Mask to RLE", "url": "/competitions/hubmap-kidney-segmentation/discussion/232357", "author_name": "", "post_date": "2021-04-13T11:52:47.556419900Z", "votes": 28, "comment_count": 6, "views": 0, "content": "

Below is fast code from @xhlulu to convert mask to rle from his notebook here

\n

(The original code comes from @paulorzp )

\n
def mask2rle(img):\n    '''\n    Efficient implementation of mask2rle, from @paulorzp\n    --\n    img: numpy array, 1 - mask, 0 - background\n    Returns run length as string formated\n    Source: https://www.kaggle.com/xhlulu/efficient-mask2rle\n    '''\n    pixels = img.T.flatten()\n    pixels = np.pad(pixels, ((1, 1), ))\n    runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n    runs[1::2] -= runs[::2]\n    return ' '.join(str(x) for x in runs)\n
\n

Example usage is below

\n
mask = np.array([[1,0,1,0],\n                 [1,1,0,1],\n                 [1,1,0,0]])\nrle = mask2rle(x)\nprint( rle )\n
\n

this outputs the string

\n
'1 3 5 3 11 1'\n
", "messages": [ { "id": "1272306", "postDate": "04/13/2021 11:52:47", "content": "

Below is fast code from @xhlulu to convert mask to rle from his notebook here

\n

(The original code comes from @paulorzp )

\n
def mask2rle(img):\n    '''\n    Efficient implementation of mask2rle, from @paulorzp\n    --\n    img: numpy array, 1 - mask, 0 - background\n    Returns run length as string formated\n    Source: https://www.kaggle.com/xhlulu/efficient-mask2rle\n    '''\n    pixels = img.T.flatten()\n    pixels = np.pad(pixels, ((1, 1), ))\n    runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n    runs[1::2] -= runs[::2]\n    return ' '.join(str(x) for x in runs)\n
\n

Example usage is below

\n
mask = np.array([[1,0,1,0],\n                 [1,1,0,1],\n                 [1,1,0,0]])\nrle = mask2rle(x)\nprint( rle )\n
\n

this outputs the string

\n
'1 3 5 3 11 1'\n
", "rawMarkdown": "Below is fast code from @xhlulu to convert `mask` to `rle` from his notebook [here][1] \n\n(The original code comes from @paulorzp )\n\n def mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n Source: https://www.kaggle.com/xhlulu/efficient-mask2rle\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\n\nExample usage is below\n\n mask = np.array([[1,0,1,0],\n [1,1,0,1],\n [1,1,0,0]])\n rle = mask2rle(x)\n print( rle )\n\nthis outputs the string\n\n '1 3 5 3 11 1'\n\n[1]: https://www.kaggle.com/xhlulu/efficient-mask2rle", "votes": null }, { "id": "1272470", "postDate": "04/13/2021 13:56:32", "content": "

very nice! thx!

", "rawMarkdown": "very nice! thx!", "votes": null }, { "id": "1276346", "postDate": "04/17/2021 12:36:33", "content": "

From the mask2rle docstring:

\n
\n

Efficient implementation of mask2rle, from @paulorzp

\n
\n

The lineage of authors is complete I guess now. 😄

", "rawMarkdown": "From the `mask2rle` docstring: \n\n> Efficient implementation of mask2rle, from @paulorzp \n\nThe lineage of authors is complete I guess now. 😄", "votes": null }, { "id": "1277557", "postDate": "04/18/2021 22:57:08", "content": "

Thanks, i updated post to indicate original author

", "rawMarkdown": "Thanks, i updated post to indicate original author", "votes": null }, { "id": "1279341", "postDate": "04/20/2021 19:41:01", "content": "

Hi @cdeotte, any idea why +1 in np.where(pixels[1:] != pixels[:-1])[0] + 1?

", "rawMarkdown": "Hi @cdeotte, any idea why `+1` in `np.where(pixels[1:] != pixels[:-1])[0] + 1`?", "votes": null }, { "id": "1279429", "postDate": "04/20/2021 22:41:18", "content": "

Yes. The first pixel in the top left corner is pixel number 1. It does not start at 0. For example the array

\n
mask = np.array([[1,0,0,0],\n                 [1,0,0,0],\n                 [0,0,0,0]])\nrle = mask2rle(x)\nprint( rle )\n
\n

should print the rle string \"1 2\" not \"0 2\".

", "rawMarkdown": "Yes. The first pixel in the top left corner is pixel number 1. It does not start at 0. For example the array\n\n mask = np.array([[1,0,0,0],\n [1,0,0,0],\n [0,0,0,0]])\n rle = mask2rle(x)\n print( rle )\n\nshould print the rle string `\"1 2\"` not `\"0 2\"`.", "votes": null }, { "id": "1282033", "postDate": "04/23/2021 14:24:29", "content": "

thank you for your sharing

", "rawMarkdown": "thank you for your sharing", "votes": null } ], "comments": [ { "id": 1272470, "author_name": "gwanghan", "author_url": "", "post_date": "04/13/2021 13:56:32", "content": "

very nice! thx!

", "votes": null, "replies": [] }, { "id": 1276346, "author_name": "yassinealouini", "author_url": "", "post_date": "04/17/2021 12:36:33", "content": "

From the mask2rle docstring:

\n
\n

Efficient implementation of mask2rle, from @paulorzp

\n
\n

The lineage of authors is complete I guess now. 😄

", "votes": null, "replies": [ { "id": 1277557, "author_name": "cdeotte", "author_url": "", "post_date": "04/18/2021 22:57:08", "content": "

Thanks, i updated post to indicate original author

", "votes": null, "replies": [] } ] }, { "id": 1279341, "author_name": "awsaf49", "author_url": "", "post_date": "04/20/2021 19:41:01", "content": "

Hi @cdeotte, any idea why +1 in np.where(pixels[1:] != pixels[:-1])[0] + 1?

", "votes": null, "replies": [ { "id": 1279429, "author_name": "cdeotte", "author_url": "", "post_date": "04/20/2021 22:41:18", "content": "

Yes. The first pixel in the top left corner is pixel number 1. It does not start at 0. For example the array

\n
mask = np.array([[1,0,0,0],\n                 [1,0,0,0],\n                 [0,0,0,0]])\nrle = mask2rle(x)\nprint( rle )\n
\n

should print the rle string \"1 2\" not \"0 2\".

", "votes": null, "replies": [] } ] }, { "id": 1282033, "author_name": "yeahbrian", "author_url": "", "post_date": "04/23/2021 14:24:29", "content": "

thank you for your sharing

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1272306": "Below is fast code from @xhlulu to convert `mask` to `rle` from his notebook [here][1] \n\n(The original code comes from @paulorzp )\n\n def mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n Source: https://www.kaggle.com/xhlulu/efficient-mask2rle\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\n\nExample usage is below\n\n mask = np.array([[1,0,1,0],\n [1,1,0,1],\n [1,1,0,0]])\n rle = mask2rle(x)\n print( rle )\n\nthis outputs the string\n\n '1 3 5 3 11 1'\n\n[1]: https://www.kaggle.com/xhlulu/efficient-mask2rle", "1272470": "very nice! thx!", "1276346": "From the `mask2rle` docstring: \n\n> Efficient implementation of mask2rle, from @paulorzp \n\nThe lineage of authors is complete I guess now. 😄", "1277557": "Thanks, i updated post to indicate original author", "1279341": "Hi @cdeotte, any idea why `+1` in `np.where(pixels[1:] != pixels[:-1])[0] + 1`?", "1279429": "Yes. The first pixel in the top left corner is pixel number 1. It does not start at 0. For example the array\n\n mask = np.array([[1,0,0,0],\n [1,0,0,0],\n [0,0,0,0]])\n rle = mask2rle(x)\n print( rle )\n\nshould print the rle string `\"1 2\"` not `\"0 2\"`.", "1282033": "thank you for your sharing" }, "source": "meta" }