{ "id": 221405, "title": "need clarification on this competition", "url": "/competitions/hpa-single-cell-image-classification/discussion/221405", "author_name": "IamSparky", "post_date": "2021-02-22T16:00:17.782000", "votes": 2, "comment_count": 2, "views": 0, "content": "

There are couple of questions I would like put for my fellow kagglers regarding this competition :-->

\n
    \n

  1. Is this a segmentation problem or a classification problem ?

    2. \n

  2. What is prediction string present in the submission & how to generate it while doing prediction with the model we will train?

    3. \n

  3. How to write the function for the metric required for the competition?

    4. \n
\n

Early thanks if you reply onto this topic.

\n

Cheers

", "messages": [ { "id": 1214267, "postDate": "2021-02-22T17:36:40.773Z", "content": "
    \n
  1. I would call weakly-labeled instance segmentation
    2. \n
  2. https://www.kaggle.com/c/hpa-single-cell-image-classification/overview/evaluation is how you generate the prediction string from your masks
    3. \n
  3. for understanding map better, check out this notebook: https://www.kaggle.com/its7171/map-understanding-with-code-and-its-tips?sort=recent-comments
    4. \n
", "rawMarkdown": "1. I would call weakly-labeled instance segmentation\n2. https://www.kaggle.com/c/hpa-single-cell-image-classification/overview/evaluation is how you generate the prediction string from your masks\n3. for understanding map better, check out this notebook: https://www.kaggle.com/its7171/map-understanding-with-code-and-its-tips?sort=recent-comments", "votes": 2, "replies": [ { "id": 1214755, "postDate": "2021-02-23T05:26:25.457Z", "content": "

Just adding on point 3, very likely you won't be able to use that evaluation function (mAP) in your code, because we don't have the instance segmentation targets in the train set. So you need to work with some proxy (e.g. classification metrics) and count on the public LB to be representative of the hidden test set :)

", "rawMarkdown": "Just adding on point 3, very likely you won't be able to use that evaluation function (mAP) in your code, because we don't have the instance segmentation targets in the train set. So you need to work with some proxy (e.g. classification metrics) and count on the public LB to be representative of the hidden test set :)", "votes": 2 } ] }, { "id": 1214129, "postDate": "2021-02-22T16:00:17.783Z", "content": "

There are couple of questions I would like put for my fellow kagglers regarding this competition :-->

\n
    \n

  1. Is this a segmentation problem or a classification problem ?

    2. \n

  2. What is prediction string present in the submission & how to generate it while doing prediction with the model we will train?

    3. \n

  3. How to write the function for the metric required for the competition?

    4. \n
\n

Early thanks if you reply onto this topic.

\n

Cheers

", "rawMarkdown": "There are couple of questions I would like put for my fellow kagglers regarding this competition :-->\n\n1. Is this a segmentation problem or a classification problem ?\n\n2. What is prediction string present in the submission & how to generate it while doing prediction with the model we will train?\n\n3. How to write the function for the metric required for the competition? \n\nEarly thanks if you reply onto this topic.\n\nCheers", "votes": 2 } ], "comments": [ { "id": 1214267, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-02-22T17:36:40.773000", "content": "
    \n
  1. I would call weakly-labeled instance segmentation
    2. \n
  2. https://www.kaggle.com/c/hpa-single-cell-image-classification/overview/evaluation is how you generate the prediction string from your masks
    3. \n
  3. for understanding map better, check out this notebook: https://www.kaggle.com/its7171/map-understanding-with-code-and-its-tips?sort=recent-comments
    4. \n
", "votes": 2, "replies": [ { "id": 1214755, "author_name": "Darek Kłeczek", "author_url": "", "post_date": "2021-02-23T05:26:25.457000", "content": "

Just adding on point 3, very likely you won't be able to use that evaluation function (mAP) in your code, because we don't have the instance segmentation targets in the train set. So you need to work with some proxy (e.g. classification metrics) and count on the public LB to be representative of the hidden test set :)

", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "1214267": "1. I would call weakly-labeled instance segmentation\n2. https://www.kaggle.com/c/hpa-single-cell-image-classification/overview/evaluation is how you generate the prediction string from your masks\n3. for understanding map better, check out this notebook: https://www.kaggle.com/its7171/map-understanding-with-code-and-its-tips?sort=recent-comments", "1214129": "There are couple of questions I would like put for my fellow kagglers regarding this competition :-->\n\n1. Is this a segmentation problem or a classification problem ?\n\n2. What is prediction string present in the submission & how to generate it while doing prediction with the model we will train?\n\n3. How to write the function for the metric required for the competition? \n\nEarly thanks if you reply onto this topic.\n\nCheers" } }