![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F991320%2Fd61169b9a226716977b54fb8fb82e393%2FScreen%20Shot%202020-11-17%20at%209.58.45%20AM.png?generation=1605625140541266&alt=media\")
\nlink to the github: https://github.com/mpicbg-csbd/stardist
The classical segmentation approach can fail for biological cell task due to a crowded environment and multiple overlapping cells. Recently a new method was proposed called StarDist (https://arxiv.org/pdf/1806.03535.pdf) and splinedist (https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf). Both of these approaches rely on approximating cell shape. In StarDist this achieved by predicting cell center and 32 vectors that span from the center to the edge of the cell. The training is done using UNET but last layer is modified.
\n
\nlink to the github: https://github.com/mpicbg-csbd/stardist
Good luck everyone =)
", "messages": [ { "id": "1082066", "postDate": "11/17/2020 14:59:15", "content": "The classical segmentation approach can fail for biological cell task due to a crowded environment and multiple overlapping cells. Recently a new method was proposed called StarDist (https://arxiv.org/pdf/1806.03535.pdf) and splinedist (https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf). Both of these approaches rely on approximating cell shape. In StarDist this achieved by predicting cell center and 32 vectors that span from the center to the edge of the cell. The training is done using UNET but last layer is modified.
\n
\nlink to the github: https://github.com/mpicbg-csbd/stardist
Good luck everyone =)
", "rawMarkdown": "The classical segmentation approach can fail for biological cell task due to a crowded environment and multiple overlapping cells. Recently a new method was proposed called StarDist (https://arxiv.org/pdf/1806.03535.pdf) and splinedist (https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf). Both of these approaches rely on approximating cell shape. In StarDist this achieved by predicting cell center and 32 vectors that span from the center to the edge of the cell. The training is done using UNET but last layer is modified. \n\n\nlink to the github: https://github.com/mpicbg-csbd/stardist\n\n\nGood luck everyone =)", "votes": null }, { "id": "1082070", "postDate": "11/17/2020 15:04:06", "content": "I'm having problems with the first link, then i found this: SplineDist
", "rawMarkdown": "I'm having problems with the first link, then i found this: [SplineDist](https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf)", "votes": null }, { "id": "1083638", "postDate": "11/19/2020 07:38:27", "content": "Thank you for sharing 😊
\nIt also seems like a good idea to look for models, papers, and implementations of SOTA in medical imaging, especially in cellular imaging, at the following sites 👇
\nmedical image segmentation in paperwithcode
SOTA for the DSB2018 datasets might be the first candidate.
\nMy current focus is on Double U-Net.
\narxiv: https://arxiv.org/pdf/2006.04868.pdf
\ngithub: https://github.com/DebeshJha/2020-CBMS-DoubleU-Net
![\"pwc\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F479538%2F9cb1b577fb02511a2ca85364b180b1ea%2Fjpg.jpg?generation=1605771029260541&alt=media\")
https://www.nature.com/articles/s41592-020-01008-z
\nhttps://github.com/MIC-DKFZ/nnUNet
\nWithout manual intervention, nnU-Net surpasses most existing approaches, including highly specialized solutions on 23 public datasets used in international biomedical segmentation competitions. We make nnU-Net publicly available as an out-of-the-box tool, rendering state-of-the-art segmentation accessible to a broad audience by requiring neither expert knowledge nor computing resources beyond standard network training
i am very much interested in the performance of StarDist or SlineDist. Did anyone implement them yet?
", "rawMarkdown": "i am very much interested in the performance of StarDist or SlineDist. Did anyone implement them yet?", "votes": null } ], "comments": [ { "id": 1082070, "author_name": "hiramcho", "author_url": "", "post_date": "11/17/2020 15:04:06", "content": "I'm having problems with the first link, then i found this: SplineDist
", "votes": null, "replies": [] }, { "id": 1083638, "author_name": "maxwell110", "author_url": "", "post_date": "11/19/2020 07:38:27", "content": "Thank you for sharing 😊
\nIt also seems like a good idea to look for models, papers, and implementations of SOTA in medical imaging, especially in cellular imaging, at the following sites 👇
\nmedical image segmentation in paperwithcode
SOTA for the DSB2018 datasets might be the first candidate.
\nMy current focus is on Double U-Net.
\narxiv: https://arxiv.org/pdf/2006.04868.pdf
\ngithub: https://github.com/DebeshJha/2020-CBMS-DoubleU-Net
https://www.nature.com/articles/s41592-020-01008-z
\nhttps://github.com/MIC-DKFZ/nnUNet
\nWithout manual intervention, nnU-Net surpasses most existing approaches, including highly specialized solutions on 23 public datasets used in international biomedical segmentation competitions. We make nnU-Net publicly available as an out-of-the-box tool, rendering state-of-the-art segmentation accessible to a broad audience by requiring neither expert knowledge nor computing resources beyond standard network training
i am very much interested in the performance of StarDist or SlineDist. Did anyone implement them yet?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1082066": "The classical segmentation approach can fail for biological cell task due to a crowded environment and multiple overlapping cells. Recently a new method was proposed called StarDist (https://arxiv.org/pdf/1806.03535.pdf) and splinedist (https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf). Both of these approaches rely on approximating cell shape. In StarDist this achieved by predicting cell center and 32 vectors that span from the center to the edge of the cell. The training is done using UNET but last layer is modified. \n\n\nlink to the github: https://github.com/mpicbg-csbd/stardist\n\n\nGood luck everyone =)", "1082070": "I'm having problems with the first link, then i found this: [SplineDist](https://www.biorxiv.org/content/10.1101/2020.10.27.357640v1.full.pdf)", "1083638": "Thank you for sharing 😊\nIt also seems like a good idea to look for models, papers, and implementations of SOTA in medical imaging, especially in cellular imaging, at the following sites 👇\n[medical image segmentation in paperwithcode](https://paperswithcode.com/task/medical-image-segmentation)\n\nSOTA for the `DSB2018` datasets might be the first candidate.\nMy current focus is on **Double U-Net**.\narxiv: [https://arxiv.org/pdf/2006.04868.pdf](https://arxiv.org/pdf/2006.04868.pdf)\ngithub: [https://github.com/DebeshJha/2020-CBMS-DoubleU-Net](https://github.com/DebeshJha/2020-CBMS-DoubleU-Net)\n\n", "1106184": "https://www.nature.com/articles/s41592-020-01008-z\nhttps://github.com/MIC-DKFZ/nnUNet\n`Without manual intervention, nnU-Net surpasses most existing approaches, including highly specialized solutions on 23 public datasets used in international biomedical segmentation competitions. We make nnU-Net publicly available as an out-of-the-box tool, rendering state-of-the-art segmentation accessible to a broad audience by requiring neither expert knowledge nor computing resources beyond standard network training`", "1109214": "i am very much interested in the performance of StarDist or SlineDist. Did anyone implement them yet?" }, "source": "meta" }