{ "id": 293033, "title": "Let Me Simplify The Problem Statement & Provide Solutions to You😊", "url": "/competitions/sartorius-cell-instance-segmentation/discussion/293033", "author_name": "", "post_date": "2021-12-04T07:57:08.354145100Z", "votes": 9, "comment_count": 1, "views": 0, "content": "

This Topic is for beginners who are starting the competition and I will Simplify the Problem Statement for You and Provide some Do's and Dont's and Give some starter Basic Material To Read that will help you gain confidence.

\n

Let's Start

\n

PROBLEM STATEMENT IN ONE LINE-
\nOur main aim is to do the Instance Segmentation of neuronal cells in Microscopic Images.

\n

Why do we have to identify those cells?
\nBecause these cells can cause diseases such as Alzheimer's and brain tumors, that is the cause of death and disability across the globe.

\n

Evaluation-
\nThis competition is evaluated on the mean average precision at the different intersections over union (IoU) thresholds.
\nMany Terms are used in Evaluation like Average Precision(AP), Mean Average Precision(mAP), threshold, Intersection Over Union(IOU), TP, FP &FN.

\n

To Understand these Terms in a clear manner do give it a read at this
\nBlog Post

\n

Some Points that are mentioned in the Competition and Need to Take Care of:

\n
    \n
  1. Current solutions have limited accuracy for neuronal cells in particular. There are 3 Types of neuronal cells present in dataset:shsy5y,astro,cort. Instance Segmentation of the neuroblastoma cell line SH-SY5Y consistently exhibits the lowest precision scores out of eight different cancer cell types tested. This could be because neuronal cells have a very unique, irregular, and concave morphology associated with them, making them challenging to segment with commonly used mask heads
    2. \n
\n\n
    \n
  1. IMP NOTE:- While predictions are not allowed to overlap, the training labels are provided in full (with overlapping portions included). This is to ensure that models are provided the full data for each object. Removing overlap in predictions is a task for the competitor.
    2. \n
\n

Now, for Instance, Segmentation Mask R-CNN is the current SOTA technique.
\nSo, here is a blog post that describes about the Instance segmentation and MASK R-CNN in a very easy & intuitive way Instance Segmentation & MASK R-CNN.
\nTo Know about the implementation details of Mask R-CNN in PyTorch look at docs

\n\n

MY NOTEBOOK EXPLAINING THE APPROACH WITH EDA

", "messages": [ { "id": "1605424", "postDate": "12/04/2021 07:57:08", "content": "

This Topic is for beginners who are starting the competition and I will Simplify the Problem Statement for You and Provide some Do's and Dont's and Give some starter Basic Material To Read that will help you gain confidence.

\n

Let's Start

\n

PROBLEM STATEMENT IN ONE LINE-
\nOur main aim is to do the Instance Segmentation of neuronal cells in Microscopic Images.

\n

Why do we have to identify those cells?
\nBecause these cells can cause diseases such as Alzheimer's and brain tumors, that is the cause of death and disability across the globe.

\n

Evaluation-
\nThis competition is evaluated on the mean average precision at the different intersections over union (IoU) thresholds.
\nMany Terms are used in Evaluation like Average Precision(AP), Mean Average Precision(mAP), threshold, Intersection Over Union(IOU), TP, FP &FN.

\n

To Understand these Terms in a clear manner do give it a read at this
\nBlog Post

\n

Some Points that are mentioned in the Competition and Need to Take Care of:

\n
    \n
  1. Current solutions have limited accuracy for neuronal cells in particular. There are 3 Types of neuronal cells present in dataset:shsy5y,astro,cort. Instance Segmentation of the neuroblastoma cell line SH-SY5Y consistently exhibits the lowest precision scores out of eight different cancer cell types tested. This could be because neuronal cells have a very unique, irregular, and concave morphology associated with them, making them challenging to segment with commonly used mask heads
    2. \n
\n\n
    \n
  1. IMP NOTE:- While predictions are not allowed to overlap, the training labels are provided in full (with overlapping portions included). This is to ensure that models are provided the full data for each object. Removing overlap in predictions is a task for the competitor.
    2. \n
\n

Now, for Instance, Segmentation Mask R-CNN is the current SOTA technique.
\nSo, here is a blog post that describes about the Instance segmentation and MASK R-CNN in a very easy & intuitive way Instance Segmentation & MASK R-CNN.
\nTo Know about the implementation details of Mask R-CNN in PyTorch look at docs

\n\n

MY NOTEBOOK EXPLAINING THE APPROACH WITH EDA

", "rawMarkdown": "This Topic is for beginners who are starting the competition and I will Simplify the Problem Statement for You and Provide some Do's and Dont's and Give some starter Basic Material To Read that will help you gain confidence.\n\n**Let's Start**\n\n**PROBLEM STATEMENT IN ONE LINE-** \nOur main aim is to do the *Instance Segmentation* of neuronal cells in Microscopic Images.\n\n**Why do we have to identify those cells?**\nBecause these cells can cause diseases such as Alzheimer's and brain tumors, that is the cause of death and disability across the globe.\n\n**Evaluation-**\nThis competition is evaluated on the mean average precision at the different intersections over union (IoU) thresholds.\n`Many Terms are used in Evaluation like Average Precision(AP), Mean Average Precision(mAP), threshold, Intersection Over Union(IOU), TP, FP &FN.`\n\nTo Understand these Terms in a clear manner do give it a read at this \n[Blog Post](https://yanfengliux.medium.com/the-confusing-metrics-of-ap-and-map-for-object-detection-3113ba0386ef)\n\n**Some Points that are mentioned in the Competition and Need to Take Care of:**\n1. Current solutions have limited accuracy for neuronal cells in particular. There are 3 Types of neuronal cells present in dataset:`shsy5y`,`astro`,`cort`. Instance Segmentation of the neuroblastoma cell line SH-SY5Y consistently exhibits the lowest precision scores out of eight different cancer cell types tested. This could be because neuronal cells have a very unique, irregular, and concave morphology associated with them, making them challenging to segment with commonly used mask heads\n* `So, you can specifically give attention to these type of cells to improve the score`.\n* ***NOTE:*** Extra LIVECell_dataset_2021 was also given which can improve your score .LIVECell is the predecessor dataset to this competition. You will find extra data for the SH-SHY5Y cell line, plus several other cell lines not covered in the competition dataset that may be of interest for transfer learning.\n\n2. **IMP NOTE:-** While predictions are not allowed to overlap, the training labels are provided in full (with overlapping portions included). This is to ensure that models are provided the full data for each object. Removing overlap in predictions is a task for the competitor.\n\nNow, for Instance, Segmentation Mask R-CNN is the current SOTA technique.\nSo, here is a blog post that describes about the Instance segmentation and MASK R-CNN in a very easy & intuitive way [Instance Segmentation & MASK R-CNN](https://viso.ai/deep-learning/mask-r-cnn/).\nTo Know about the implementation details of Mask R-CNN in PyTorch look at [docs](https://pytorch.org/vision/stable/models.html#object-detection-instance-segmentation-and-person-keypoint-detection)\n\n* Another Resource used by highest voted notebook [TORCHVISION INSTANCE SEGMENTATION FINETUNING TUTORIAL](https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html)\n\n[MY NOTEBOOK EXPLAINING THE APPROACH WITH EDA](https://www.kaggle.com/harshwalia/sartorius-implementation-guide-for-mask-rcnn-eda)", "votes": null }, { "id": "1633524", "postDate": "12/30/2021 21:27:11", "content": "

Annotations have rle representataion and also cell type . What should be the output prediction.

", "rawMarkdown": "Annotations have rle representataion and also cell type . What should be the output prediction.", "votes": null } ], "comments": [ { "id": 1633524, "author_name": "bogivinaykumar", "author_url": "", "post_date": "12/30/2021 21:27:11", "content": "

Annotations have rle representataion and also cell type . What should be the output prediction.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1605424": "This Topic is for beginners who are starting the competition and I will Simplify the Problem Statement for You and Provide some Do's and Dont's and Give some starter Basic Material To Read that will help you gain confidence.\n\n**Let's Start**\n\n**PROBLEM STATEMENT IN ONE LINE-** \nOur main aim is to do the *Instance Segmentation* of neuronal cells in Microscopic Images.\n\n**Why do we have to identify those cells?**\nBecause these cells can cause diseases such as Alzheimer's and brain tumors, that is the cause of death and disability across the globe.\n\n**Evaluation-**\nThis competition is evaluated on the mean average precision at the different intersections over union (IoU) thresholds.\n`Many Terms are used in Evaluation like Average Precision(AP), Mean Average Precision(mAP), threshold, Intersection Over Union(IOU), TP, FP &FN.`\n\nTo Understand these Terms in a clear manner do give it a read at this \n[Blog Post](https://yanfengliux.medium.com/the-confusing-metrics-of-ap-and-map-for-object-detection-3113ba0386ef)\n\n**Some Points that are mentioned in the Competition and Need to Take Care of:**\n1. Current solutions have limited accuracy for neuronal cells in particular. There are 3 Types of neuronal cells present in dataset:`shsy5y`,`astro`,`cort`. Instance Segmentation of the neuroblastoma cell line SH-SY5Y consistently exhibits the lowest precision scores out of eight different cancer cell types tested. This could be because neuronal cells have a very unique, irregular, and concave morphology associated with them, making them challenging to segment with commonly used mask heads\n* `So, you can specifically give attention to these type of cells to improve the score`.\n* ***NOTE:*** Extra LIVECell_dataset_2021 was also given which can improve your score .LIVECell is the predecessor dataset to this competition. You will find extra data for the SH-SHY5Y cell line, plus several other cell lines not covered in the competition dataset that may be of interest for transfer learning.\n\n2. **IMP NOTE:-** While predictions are not allowed to overlap, the training labels are provided in full (with overlapping portions included). This is to ensure that models are provided the full data for each object. Removing overlap in predictions is a task for the competitor.\n\nNow, for Instance, Segmentation Mask R-CNN is the current SOTA technique.\nSo, here is a blog post that describes about the Instance segmentation and MASK R-CNN in a very easy & intuitive way [Instance Segmentation & MASK R-CNN](https://viso.ai/deep-learning/mask-r-cnn/).\nTo Know about the implementation details of Mask R-CNN in PyTorch look at [docs](https://pytorch.org/vision/stable/models.html#object-detection-instance-segmentation-and-person-keypoint-detection)\n\n* Another Resource used by highest voted notebook [TORCHVISION INSTANCE SEGMENTATION FINETUNING TUTORIAL](https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html)\n\n[MY NOTEBOOK EXPLAINING THE APPROACH WITH EDA](https://www.kaggle.com/harshwalia/sartorius-implementation-guide-for-mask-rcnn-eda)", "1633524": "Annotations have rle representataion and also cell type . What should be the output prediction." }, "source": "meta" }