![\"img_with_colored_masks\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F172860%2F2c9fcff3f973406f5fd751fe3b1dbfd6%2FScreenshot%20from%202022-07-10%2012-05-18.png?generation=1657447799356700&alt=media&width=100\")
Another useful code snippet using skimage.color.label2rgb.
This is useful function to have both the image and the masks as a colored overlay.
\nHere is a code snippet on how to do it:
\nfrom skimage.color import label2rgb\nimport pandas as pd\nfrom PIL import Image\n\n# The same old rle2mask function, very handy!\ndef rle2mask(mask_rle, shape=(256, 256)):\n '''\n mask_rle: run-length as string formated (start length)\n shape: (width,height) of array to return \n Returns numpy array, 1 - mask, 0 - background\n Source: https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n '''\n s = mask_rle.split()\n # print(\"The length of the RLE is:\", len(s))\n starts, lengths = [np.asarray(x, dtype=int) for x in (s[0:][::2], s[1:][::2])]\n starts -= 1\n ends = starts + lengths\n img = np.zeros(shape[0]*shape[1], dtype=np.uint8)\n for lo, hi in zip(starts, ends):\n img[lo:hi] = 1\n return img.reshape(shape).T\n\n\n# Select the image id you want to display\nimg_id = 10274\nimg_path = f\"../input/hubmap-organ-segmentation/train_images/{img_id}.tiff\"\n# We also need the metadata to get the RLE.\ntrain_path = \"../input/hubmap-organ-segmentation/train.csv\"\n\npil_img = Image.open(img_path)\n# The image should be an array for the label2rgb\n# function\nimg = np.asarray(pil_img)\n\n# We get the mask from the metadata DataFrame.\ntrain_df = pd.read_csv(train_path)\nrle = train_df.loc[lambda df: df[\"id\"] == img_id, \"rle\"].values[0]\nmask = rle2mask(rle, shape=(3000, 3000))\n\n# Play a bit with the bg_color and alpha to get different renderings.\n# By default, the first mask will be in red unless you specify a different\n# list of colors.\n# If you choose bg_label=1, the background will be red.\nimg_with_colored_masks = label2rgb(mask, img, bg_label=0, \n bg_color=(1.,1.,1.), \n alpha=0.5)\nplt.imshow(img_with_colored_masks)\nHere is what the result might look like:
\n
I hope this is useful for exploring the masks.
", "messages": [ { "id": "1850314", "postDate": "07/10/2022 10:10:13", "content": "Another useful code snippet using skimage.color.label2rgb.
This is useful function to have both the image and the masks as a colored overlay.
\nHere is a code snippet on how to do it:
\nfrom skimage.color import label2rgb\nimport pandas as pd\nfrom PIL import Image\n\n# The same old rle2mask function, very handy!\ndef rle2mask(mask_rle, shape=(256, 256)):\n '''\n mask_rle: run-length as string formated (start length)\n shape: (width,height) of array to return \n Returns numpy array, 1 - mask, 0 - background\n Source: https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n '''\n s = mask_rle.split()\n # print(\"The length of the RLE is:\", len(s))\n starts, lengths = [np.asarray(x, dtype=int) for x in (s[0:][::2], s[1:][::2])]\n starts -= 1\n ends = starts + lengths\n img = np.zeros(shape[0]*shape[1], dtype=np.uint8)\n for lo, hi in zip(starts, ends):\n img[lo:hi] = 1\n return img.reshape(shape).T\n\n\n# Select the image id you want to display\nimg_id = 10274\nimg_path = f\"../input/hubmap-organ-segmentation/train_images/{img_id}.tiff\"\n# We also need the metadata to get the RLE.\ntrain_path = \"../input/hubmap-organ-segmentation/train.csv\"\n\npil_img = Image.open(img_path)\n# The image should be an array for the label2rgb\n# function\nimg = np.asarray(pil_img)\n\n# We get the mask from the metadata DataFrame.\ntrain_df = pd.read_csv(train_path)\nrle = train_df.loc[lambda df: df[\"id\"] == img_id, \"rle\"].values[0]\nmask = rle2mask(rle, shape=(3000, 3000))\n\n# Play a bit with the bg_color and alpha to get different renderings.\n# By default, the first mask will be in red unless you specify a different\n# list of colors.\n# If you choose bg_label=1, the background will be red.\nimg_with_colored_masks = label2rgb(mask, img, bg_label=0, \n bg_color=(1.,1.,1.), \n alpha=0.5)\nplt.imshow(img_with_colored_masks)\nHere is what the result might look like:
\n
I hope this is useful for exploring the masks.
", "rawMarkdown": "Another useful code snippet using [`skimage.color.label2rgb`](https://scikit-image.org/docs/dev/api/skimage.color.html#skimage.color.label2rgb). \n\nThis is useful function to have **both the image and the masks as a colored overlay**. \n\nHere is a code snippet on how to do it:\n\n\n```\nfrom skimage.color import label2rgb\nimport pandas as pd\nfrom PIL import Image\n\n# The same old rle2mask function, very handy!\ndef rle2mask(mask_rle, shape=(256, 256)):\n '''\n mask_rle: run-length as string formated (start length)\n shape: (width,height) of array to return \n Returns numpy array, 1 - mask, 0 - background\n Source: https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n '''\n s = mask_rle.split()\n # print(\"The length of the RLE is:\", len(s))\n starts, lengths = [np.asarray(x, dtype=int) for x in (s[0:][::2], s[1:][::2])]\n starts -= 1\n ends = starts + lengths\n img = np.zeros(shape[0]*shape[1], dtype=np.uint8)\n for lo, hi in zip(starts, ends):\n img[lo:hi] = 1\n return img.reshape(shape).T\n\n\n# Select the image id you want to display\nimg_id = 10274\nimg_path = f\"../input/hubmap-organ-segmentation/train_images/{img_id}.tiff\"\n# We also need the metadata to get the RLE.\ntrain_path = \"../input/hubmap-organ-segmentation/train.csv\"\n\npil_img = Image.open(img_path)\n# The image should be an array for the label2rgb\n# function\nimg = np.asarray(pil_img)\n\n# We get the mask from the metadata DataFrame.\ntrain_df = pd.read_csv(train_path)\nrle = train_df.loc[lambda df: df[\"id\"] == img_id, \"rle\"].values[0]\nmask = rle2mask(rle, shape=(3000, 3000))\n\n# Play a bit with the bg_color and alpha to get different renderings.\n# By default, the first mask will be in red unless you specify a different\n# list of colors.\n# If you choose bg_label=1, the background will be red.\nimg_with_colored_masks = label2rgb(mask, img, bg_label=0, \n bg_color=(1.,1.,1.), \n alpha=0.5)\nplt.imshow(img_with_colored_masks)\n\n```\n\nHere is what the result might look like:\n\n\n\nI hope this is useful for exploring the masks.", "votes": null }, { "id": "1850320", "postDate": "07/10/2022 10:19:02", "content": "thanks for sharing this! it is really helpful @yassinealouini
", "rawMarkdown": "thanks for sharing this! it is really helpful @yassinealouini", "votes": null }, { "id": "1850648", "postDate": "07/10/2022 15:17:53", "content": "Thanks @subhajitbag. Glad you like it. 👌
", "rawMarkdown": "Thanks @subhajitbag. Glad you like it. 👌", "votes": null }, { "id": "1850855", "postDate": "07/10/2022 20:35:30", "content": "Thanks a lot! Have you noticed this type of masks in the data? Just wondering, if it is the issue of your script or the data itself.![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1269464%2F69b74d43e4edaf9eba229d647fb2ff53%2Fw1n_l2UyEKoyx1bon7-rE20PW2gA8oWcMwcWGq2oz7SgzPdVXoCvTtJIBn4gK4_4UhP-i8Y-819DvXWPw6AEwKzs.jpg?generation=1657485320264590&alt=media\")
Wow really interesting! It will be helpful for sure.. thanks for sharing this snippet.
", "rawMarkdown": "Wow really interesting! It will be helpful for sure.. thanks for sharing this snippet.", "votes": null }, { "id": "1851248", "postDate": "07/11/2022 05:39:10", "content": "That's an interesting artifact. Which image id is it? 👀
", "rawMarkdown": "That's an interesting artifact. Which image id is it? 👀", "votes": null }, { "id": "1851249", "postDate": "07/11/2022 05:39:27", "content": "You are welcome. 👌
", "rawMarkdown": "You are welcome. 👌", "votes": null }, { "id": "1851410", "postDate": "07/11/2022 08:20:11", "content": "I don't remember which one is this particular image, but you can look into 5317, 14396, 2344. These are quite similar
", "rawMarkdown": "I don't remember which one is this particular image, but you can look into 5317, 14396, 2344. These are quite similar", "votes": null }, { "id": "1851534", "postDate": "07/11/2022 10:32:23", "content": "I don't have that type of artifact. You probably mess something up while decoding rle masks.
", "rawMarkdown": "I don't have that type of artifact. You probably mess something up while decoding rle masks.", "votes": null }, { "id": "1851944", "postDate": "07/11/2022 17:00:53", "content": "I spotted the problem. There are images of different size from (3000, 3000).
\nSo, you should change the line
\n mask = rle2mask(rle, shape=(3000, 3000))
\ninto
\nmask = rle2mask(rle, shape=img.shape[:2])
In this case the all the masks will look normal.
", "rawMarkdown": "I spotted the problem. There are images of different size from (3000, 3000).\n\nSo, you should change the line \n `mask = rle2mask(rle, shape=(3000, 3000))`\ninto\n`mask = rle2mask(rle, shape=img.shape[:2])`\n\nIn this case the all the masks will look normal.", "votes": null }, { "id": "1853326", "postDate": "07/12/2022 20:15:31", "content": "You are right of course. Most of the images are of size (3000, 3000) but some aren't. Check this plot from my \"Some insights\" discussion:
![\"img_height_distr\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F172860%2Fc26a1688d25854c1c0e9154a7b5598ec%2Finbox_172860_0ca15f350b45010e20d710025cb1986d_img_width_distribution.png?generation=1657656920671560&alt=media\")
thanks for sharing this! it is really helpful @yassinealouini
", "votes": null, "replies": [ { "id": 1850648, "author_name": "yassinealouini", "author_url": "", "post_date": "07/10/2022 15:17:53", "content": "Thanks @subhajitbag. Glad you like it. 👌
", "votes": null, "replies": [] } ] }, { "id": 1850855, "author_name": "artichoke", "author_url": "", "post_date": "07/10/2022 20:35:30", "content": "Thanks a lot! Have you noticed this type of masks in the data? Just wondering, if it is the issue of your script or the data itself.
That's an interesting artifact. Which image id is it? 👀
", "votes": null, "replies": [] }, { "id": 1851410, "author_name": "artichoke", "author_url": "", "post_date": "07/11/2022 08:20:11", "content": "I don't remember which one is this particular image, but you can look into 5317, 14396, 2344. These are quite similar
", "votes": null, "replies": [] }, { "id": 1851534, "author_name": "gunesevitan", "author_url": "", "post_date": "07/11/2022 10:32:23", "content": "I don't have that type of artifact. You probably mess something up while decoding rle masks.
", "votes": null, "replies": [] }, { "id": 1851944, "author_name": "artichoke", "author_url": "", "post_date": "07/11/2022 17:00:53", "content": "I spotted the problem. There are images of different size from (3000, 3000).
\nSo, you should change the line
\n mask = rle2mask(rle, shape=(3000, 3000))
\ninto
\nmask = rle2mask(rle, shape=img.shape[:2])
In this case the all the masks will look normal.
", "votes": null, "replies": [] }, { "id": 1853326, "author_name": "yassinealouini", "author_url": "", "post_date": "07/12/2022 20:15:31", "content": "You are right of course. Most of the images are of size (3000, 3000) but some aren't. Check this plot from my \"Some insights\" discussion:
Wow really interesting! It will be helpful for sure.. thanks for sharing this snippet.
", "votes": null, "replies": [ { "id": 1851249, "author_name": "yassinealouini", "author_url": "", "post_date": "07/11/2022 05:39:27", "content": "You are welcome. 👌
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1850314": "Another useful code snippet using [`skimage.color.label2rgb`](https://scikit-image.org/docs/dev/api/skimage.color.html#skimage.color.label2rgb). \n\nThis is useful function to have **both the image and the masks as a colored overlay**. \n\nHere is a code snippet on how to do it:\n\n\n```\nfrom skimage.color import label2rgb\nimport pandas as pd\nfrom PIL import Image\n\n# The same old rle2mask function, very handy!\ndef rle2mask(mask_rle, shape=(256, 256)):\n '''\n mask_rle: run-length as string formated (start length)\n shape: (width,height) of array to return \n Returns numpy array, 1 - mask, 0 - background\n Source: https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n '''\n s = mask_rle.split()\n # print(\"The length of the RLE is:\", len(s))\n starts, lengths = [np.asarray(x, dtype=int) for x in (s[0:][::2], s[1:][::2])]\n starts -= 1\n ends = starts + lengths\n img = np.zeros(shape[0]*shape[1], dtype=np.uint8)\n for lo, hi in zip(starts, ends):\n img[lo:hi] = 1\n return img.reshape(shape).T\n\n\n# Select the image id you want to display\nimg_id = 10274\nimg_path = f\"../input/hubmap-organ-segmentation/train_images/{img_id}.tiff\"\n# We also need the metadata to get the RLE.\ntrain_path = \"../input/hubmap-organ-segmentation/train.csv\"\n\npil_img = Image.open(img_path)\n# The image should be an array for the label2rgb\n# function\nimg = np.asarray(pil_img)\n\n# We get the mask from the metadata DataFrame.\ntrain_df = pd.read_csv(train_path)\nrle = train_df.loc[lambda df: df[\"id\"] == img_id, \"rle\"].values[0]\nmask = rle2mask(rle, shape=(3000, 3000))\n\n# Play a bit with the bg_color and alpha to get different renderings.\n# By default, the first mask will be in red unless you specify a different\n# list of colors.\n# If you choose bg_label=1, the background will be red.\nimg_with_colored_masks = label2rgb(mask, img, bg_label=0, \n bg_color=(1.,1.,1.), \n alpha=0.5)\nplt.imshow(img_with_colored_masks)\n\n```\n\nHere is what the result might look like:\n\n\n\nI hope this is useful for exploring the masks.", "1850320": "thanks for sharing this! it is really helpful @yassinealouini", "1850648": "Thanks @subhajitbag. Glad you like it. 👌", "1850855": "Thanks a lot! Have you noticed this type of masks in the data? Just wondering, if it is the issue of your script or the data itself.", "1850926": "Wow really interesting! It will be helpful for sure.. thanks for sharing this snippet.", "1851248": "That's an interesting artifact. Which image id is it? 👀", "1851249": "You are welcome. 👌", "1851410": "I don't remember which one is this particular image, but you can look into 5317, 14396, 2344. These are quite similar", "1851534": "I don't have that type of artifact. You probably mess something up while decoding rle masks.", "1851944": "I spotted the problem. There are images of different size from (3000, 3000).\n\nSo, you should change the line \n `mask = rle2mask(rle, shape=(3000, 3000))`\ninto\n`mask = rle2mask(rle, shape=img.shape[:2])`\n\nIn this case the all the masks will look normal.", "1853326": "You are right of course. Most of the images are of size `(3000, 3000)` but some aren't. Check this plot from my \"Some insights\" [discussion](https://www.kaggle.com/competitions/hubmap-organ-segmentation/discussion/333389):\n\n\n\n" }, "source": "meta" }