On the official Cellprofiling Github we have some pieces of code that could, potentially, be useful for this competition.
\nThe code for the solutions from the previous challenge that was later published in Nature Methods. You will also be able to get the trained weights from the Bestfitting's winning solution there.
Code for segmenting individual cells and nuclei in the HPA images.
Here are a couple that are not from our lab, but could still be useful.
\nImageJ: A great tool to explore microscopy images with. ImageJ is also available as FIJI and as a web app on Imjoy.
On the official Cellprofiling Github we have some pieces of code that could, potentially, be useful for this competition.
\nThe code for the solutions from the previous challenge that was later published in Nature Methods. You will also be able to get the trained weights from the Bestfitting's winning solution there.
Code for segmenting individual cells and nuclei in the HPA images.
Here are a couple that are not from our lab, but could still be useful.
\nImageJ: A great tool to explore microscopy images with. ImageJ is also available as FIJI and as a web app on Imjoy.
Just tried the code for segmenting individual cells. Got the following error:
\nImportError: cannot import name 'isStringType' from 'PIL._util' (/opt/conda/lib/python3.7/site-packages/PIL/_util.py)\nMore on that in this github issue where I also shared the call stack.
", "rawMarkdown": "Just tried the code for segmenting individual cells. Got the following error:\n```\nImportError: cannot import name 'isStringType' from 'PIL._util' (/opt/conda/lib/python3.7/site-packages/PIL/_util.py)\n```\nMore on that in [this github issue](https://github.com/CellProfiling/HPA-Cell-Segmentation/issues/15) where I also shared the call stack.", "replies": [ { "id": 1173144, "postDate": "2021-01-27T17:34:45.203Z", "content": "Hi! Have you try looking at this example? https://www.kaggle.com/lnhtrang/hpa-public-data-download-and-hpacellseg
\nI tried it last week with
\n!pip install https://github.com/CellProfiling/HPA-Cell-Segmentation/archive/master.zip
Thanks for the quick reply, that worked. The trick was to install the HPA Cell Segmentation package with
\n!pip install https://github.com/CellProfiling/HPA-Cell-Segmentation/archive/master.zip\nInstead of what I had locally, which is
\n!git clone https://github.com/CellProfiling/HPA-Cell-Segmentation\n!cd HPA-Cell-Segmentation; sh install.sh\nThat is odd. I have tested the sh install.sh before and had it work fine. I will look into it tomorrow. Thank you for letting us know.
I have a question that I might have missed somewhere but I just tried out the cell segmenting on some of the test images and I ran into issues with shapes other than (2048, 2048, 3) so I am thinking images outside that size need to be resized before we can create segmentations?
", "rawMarkdown": "I have a question that I might have missed somewhere but I just tried out the cell segmenting on some of the test images and I ran into issues with shapes other than (2048, 2048, 3) so I am thinking images outside that size need to be resized before we can create segmentations?" }, { "id": 1173441, "postDate": "2021-01-27T22:05:28.423Z", "content": "It should work for different sizes, however it will only work if all the images in the stack are the same size.
", "rawMarkdown": "It should work for different sizes, however it will only work if all the images in the stack are the same size.", "votes": 2 }, { "id": 1177049, "postDate": "2021-01-30T02:25:20.010Z", "content": "Hey @cwinsnes maybe I am missing something here but when I use a single image that does not have width and height of 2048 then I get this error:
\n/opt/conda/lib/python3.7/site-packages/pytorch_zoo/abstract_model.py in forward(self, dec, enc)\n 109 \n 110 def forward(self, dec, enc):\n--> 111 x = torch.cat([dec, enc], dim=1)\n 112 return self.seq(x)\n 113 \n\nRuntimeError: Sizes of tensors must match except in dimension 1. Got 28 and 27 in dimension 2 (The offending index is 1)\nIt looks like the decoder and encoder have different values in dimension 2 when I look through the debugger.
\nThis is the input
\n[['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_red.png'],\n ['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_yellow.png'],\n ['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_blue.png']]\n@rdizzl3 try to set padding=True in the arguments to cell segmentator model, it has worked for me.
", "rawMarkdown": "@rdizzl3 try to set padding=True in the arguments to cell segmentator model, it has worked for me. ", "votes": 7 }, { "id": 1177056, "postDate": "2021-01-30T02:45:02.377Z", "content": "@thedrcat I wish could give you at least 10 upvotes for this. That is what I was missing. Btw what I am working on you just published :)
", "rawMarkdown": "@thedrcat I wish could give you at least 10 upvotes for this. That is what I was missing. Btw what I am working on you just published :)" } ] } ], "comments": [ { "id": 1173113, "author_name": "Chan Kha Vu", "author_url": "", "post_date": "2021-01-27T17:22:39.910000", "content": "Just tried the code for segmenting individual cells. Got the following error:
\nImportError: cannot import name 'isStringType' from 'PIL._util' (/opt/conda/lib/python3.7/site-packages/PIL/_util.py)\nMore on that in this github issue where I also shared the call stack.
", "votes": 0, "replies": [ { "id": 1173144, "author_name": "Trang Le", "author_url": "", "post_date": "2021-01-27T17:34:45.203000", "content": "Hi! Have you try looking at this example? https://www.kaggle.com/lnhtrang/hpa-public-data-download-and-hpacellseg
\nI tried it last week with
\n!pip install https://github.com/CellProfiling/HPA-Cell-Segmentation/archive/master.zip
Thanks for the quick reply, that worked. The trick was to install the HPA Cell Segmentation package with
\n!pip install https://github.com/CellProfiling/HPA-Cell-Segmentation/archive/master.zip\nInstead of what I had locally, which is
\n!git clone https://github.com/CellProfiling/HPA-Cell-Segmentation\n!cd HPA-Cell-Segmentation; sh install.sh\nThat is odd. I have tested the sh install.sh before and had it work fine. I will look into it tomorrow. Thank you for letting us know.
I have a question that I might have missed somewhere but I just tried out the cell segmenting on some of the test images and I ran into issues with shapes other than (2048, 2048, 3) so I am thinking images outside that size need to be resized before we can create segmentations?
", "votes": 0, "replies": [] }, { "id": 1173441, "author_name": "Casper Winsnes", "author_url": "", "post_date": "2021-01-27T22:05:28.423000", "content": "It should work for different sizes, however it will only work if all the images in the stack are the same size.
", "votes": 2, "replies": [] }, { "id": 1177049, "author_name": "RDizzl3", "author_url": "", "post_date": "2021-01-30T02:25:20.010000", "content": "Hey @cwinsnes maybe I am missing something here but when I use a single image that does not have width and height of 2048 then I get this error:
\n/opt/conda/lib/python3.7/site-packages/pytorch_zoo/abstract_model.py in forward(self, dec, enc)\n 109 \n 110 def forward(self, dec, enc):\n--> 111 x = torch.cat([dec, enc], dim=1)\n 112 return self.seq(x)\n 113 \n\nRuntimeError: Sizes of tensors must match except in dimension 1. Got 28 and 27 in dimension 2 (The offending index is 1)\nIt looks like the decoder and encoder have different values in dimension 2 when I look through the debugger.
\nThis is the input
\n[['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_red.png'],\n ['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_yellow.png'],\n ['/kaggle/input/hpa-single-cell-image-classification/test/020a29cf-2c24-478b-8603-c22a90dc3e31_blue.png']]\n@rdizzl3 try to set padding=True in the arguments to cell segmentator model, it has worked for me.
", "votes": 7, "replies": [] }, { "id": 1177056, "author_name": "RDizzl3", "author_url": "", "post_date": "2021-01-30T02:45:02.377000", "content": "@thedrcat I wish could give you at least 10 upvotes for this. That is what I was missing. Btw what I am working on you just published :)
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1171645": "On the official Cellprofiling Github we have some pieces of code that could, potentially, be useful for this competition.\n\n1. [The code for the solutions from the previous challenge](https://github.com/CellProfiling/HPA-competition-solutions) that was later published in [Nature Methods](https://www.nature.com/articles/s41592-019-0658-6). You will also be able to get the trained weights from the Bestfitting's winning solution there.\n\n2. [Code for segmenting individual cells and nuclei in the HPA images](https://github.com/CellProfiling/HPA-Cell-Segmentation).\n\nHere are a couple that are not from our lab, but could still be useful.\n1. [Cellpose: A more general approach to cell segmentation](https://github.com/MouseLand/cellpose)\n\n3. [ImageJ: A great tool to explore microscopy images with](https://github.com/imagej/imagej). ImageJ is also available as [FIJI](https://github.com/fiji/fiji) and as a [web app on Imjoy](https://ij.imjoy.io/).", "1173113": "Just tried the code for segmenting individual cells. Got the following error:\n```\nImportError: cannot import name 'isStringType' from 'PIL._util' (/opt/conda/lib/python3.7/site-packages/PIL/_util.py)\n```\nMore on that in [this github issue](https://github.com/CellProfiling/HPA-Cell-Segmentation/issues/15) where I also shared the call stack." } }