{ "id": 311184, "title": "Full Body Annotations and cropped Dataset", "url": "/competitions/happy-whale-and-dolphin/discussion/311184", "author_name": "Jan Bre", "post_date": "2022-03-05T10:25:31.373000", "votes": 74, "comment_count": 5, "views": 0, "content": "

Hey everybody

\n

I am also releasing my 2nd dataset, where I have annotated 3300 images with bounding boxes.

\n

Instead of cropping only the backfin as in my previous dataset these bounding boxes include all of the whale. Therefore also including the 3 whale species, which were not part of the last dataset.

\n

My initial idea was to build models for full bodies and backfins only and stack them somewhere downstream in the pipeline.

\n

\"fullbody.png\"

\n

The dataset can be found here:

\n

Please upvote the dataset if you are using it
\nhttps://www.kaggle.com/jpbremer/fullbodywhaleannotations

\n

It includes:

\n
    \n

  1. the fullbodyannotations.csv file which are the ground truth files.

    2. \n

  2. train.csv + test.csv, which are retrieved using yolov5 object detection based on fullbodyannotations.csv

    3. \n

  3. two folders with the **cropped train and test images **processed using train + test.csv

    4. \n
\n

Cheers

", "messages": [ { "id": 1712780, "postDate": "2022-03-05T10:25:31.373Z", "content": "

Hey everybody

\n

I am also releasing my 2nd dataset, where I have annotated 3300 images with bounding boxes.

\n

Instead of cropping only the backfin as in my previous dataset these bounding boxes include all of the whale. Therefore also including the 3 whale species, which were not part of the last dataset.

\n

My initial idea was to build models for full bodies and backfins only and stack them somewhere downstream in the pipeline.

\n

\"fullbody.png\"

\n

The dataset can be found here:

\n

Please upvote the dataset if you are using it
\nhttps://www.kaggle.com/jpbremer/fullbodywhaleannotations

\n

It includes:

\n
    \n

  1. the fullbodyannotations.csv file which are the ground truth files.

    2. \n

  2. train.csv + test.csv, which are retrieved using yolov5 object detection based on fullbodyannotations.csv

    3. \n

  3. two folders with the **cropped train and test images **processed using train + test.csv

    4. \n
\n

Cheers

", "rawMarkdown": "**Hey everybody**\n\nI am also releasing my 2nd dataset, where I have annotated 3300 images with bounding boxes. \n\nInstead of cropping only the backfin as in my previous dataset these bounding boxes include all of the whale. Therefore also including the 3 whale species, which were not part of the last dataset.\n\nMy initial idea was to build models for full bodies and backfins only and stack them somewhere downstream in the pipeline.\n\n[![fullbody.png](https://i.postimg.cc/tTYdzYcj/fullbody.png)](https://postimg.cc/CBpfhLYP)\n\n\nThe dataset can be found here: \n\n**Please upvote the dataset if you are using it**\nhttps://www.kaggle.com/jpbremer/fullbodywhaleannotations\n\n\n\n\nIt includes:\n1. the **fullbodyannotations.csv** file which are the ground truth files. \n\n2. **train.csv + test.csv,** which are retrieved using yolov5 object detection based on fullbodyannotations.csv\n\n3. two folders with the **cropped train and test images **processed using train + test.csv\n\n**Cheers**\n", "votes": 74 }, { "id": 1712840, "postDate": "2022-03-05T12:05:47.600Z", "content": "

great! what is your LB of the dataset?

", "rawMarkdown": "great! what is your LB of the dataset?", "replies": [ { "id": 1712844, "postDate": "2022-03-05T12:09:44.367Z", "content": "

Did not use it. Dont have time unfortunately. :(

", "rawMarkdown": "Did not use it. Dont have time unfortunately. :(", "votes": 1 } ] }, { "id": 1729604, "postDate": "2022-03-20T09:21:20.853Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1732288, "postDate": "2022-03-23T07:37:03.380Z", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!" }, { "id": 1712930, "postDate": "2022-03-05T13:44:54.927Z", "content": "

Amazing ! Thank you for sharing πŸ‘

", "rawMarkdown": "Amazing ! Thank you for sharing πŸ‘" } ], "comments": [ { "id": 1712840, "author_name": "dragon zhang", "author_url": "", "post_date": "2022-03-05T12:05:47.600000", "content": "

great! what is your LB of the dataset?

", "votes": 0, "replies": [ { "id": 1712844, "author_name": "Jan Bre", "author_url": "", "post_date": "2022-03-05T12:09:44.367000", "content": "

Did not use it. Dont have time unfortunately. :(

", "votes": 1, "replies": [] } ] }, { "id": 1729604, "author_name": "", "author_url": "", "post_date": "2022-03-20T09:21:20.853000", "content": "", "votes": 0, "replies": [] }, { "id": 1732288, "author_name": "Jiahuan Cheng", "author_url": "", "post_date": "2022-03-23T07:37:03.380000", "content": "

Thanks for sharing!

", "votes": 0, "replies": [] }, { "id": 1712930, "author_name": "ThΓ©o Boyer", "author_url": "", "post_date": "2022-03-05T13:44:54.927000", "content": "

Amazing ! Thank you for sharing πŸ‘

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1712780": "**Hey everybody**\n\nI am also releasing my 2nd dataset, where I have annotated 3300 images with bounding boxes. \n\nInstead of cropping only the backfin as in my previous dataset these bounding boxes include all of the whale. Therefore also including the 3 whale species, which were not part of the last dataset.\n\nMy initial idea was to build models for full bodies and backfins only and stack them somewhere downstream in the pipeline.\n\n[![fullbody.png](https://i.postimg.cc/tTYdzYcj/fullbody.png)](https://postimg.cc/CBpfhLYP)\n\n\nThe dataset can be found here: \n\n**Please upvote the dataset if you are using it**\nhttps://www.kaggle.com/jpbremer/fullbodywhaleannotations\n\n\n\n\nIt includes:\n1. the **fullbodyannotations.csv** file which are the ground truth files. \n\n2. **train.csv + test.csv,** which are retrieved using yolov5 object detection based on fullbodyannotations.csv\n\n3. two folders with the **cropped train and test images **processed using train + test.csv\n\n**Cheers**\n", "1712840": "great! what is your LB of the dataset?", "1729604": "", "1732288": "Thanks for sharing!", "1712930": "Amazing ! Thank you for sharing πŸ‘" } }