![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/442660/10911/Capture.PNG\")
Hi everyone!\nI see many people use bounding boxes trained on what seems to be playground images. Like this kernel have a file named croppings.txt\nhttps://www.kaggle.com/martinpiotte/bounding-box-model/notebook\nBut the file has image filenames that are not present in the playground dataset. I just downloaded playground dataset and it is equal to the current competition dataset.\nThe competition host representative said \"A Playground version of this competition was hosted a few months ago. This competition contains even more images and individual whales than the last. \" But in reality these two competitions share same datasets now.\nAm I missing something?\nThanks!
", "messages": [ { "id": "442349", "postDate": "12/19/2018 21:01:36", "content": "Hi everyone!\nI see many people use bounding boxes trained on what seems to be playground images. Like this kernel have a file named croppings.txt\nhttps://www.kaggle.com/martinpiotte/bounding-box-model/notebook\nBut the file has image filenames that are not present in the playground dataset. I just downloaded playground dataset and it is equal to the current competition dataset.\nThe competition host representative said \"A Playground version of this competition was hosted a few months ago. This competition contains even more images and individual whales than the last. \" But in reality these two competitions share same datasets now.\nAm I missing something?\nThanks!
", "rawMarkdown": "Hi everyone!\nI see many people use bounding boxes trained on what seems to be playground images. Like this kernel have a file named croppings.txt\nhttps://www.kaggle.com/martinpiotte/bounding-box-model/notebook\nBut the file has image filenames that are not present in the playground dataset. I just downloaded playground dataset and it is equal to the current competition dataset.\nThe competition host representative said \"A Playground version of this competition was hosted a few months ago. This competition contains even more images and individual whales than the last. \" But in reality these two competitions share same datasets now.\nAm I missing something?\nThanks!", "votes": null }, { "id": "442623", "postDate": "12/20/2018 08:21:18", "content": "The data is still downloadable from the playground competition - one option would be to use it to train the model. On the other hand, it would be also interesting to verify there is no overlap between the train set from the previous comp and the test set here. Or if indeed all the images in the earlier train set are contained in this train set. Such analysis could be very useful.
\n\nAs for bounding boxes, I am currently in the process of creating my own annotations. Aiming for around a 1000, maybe a bit more, we'll see. Planning on sharing them but it is not clear how long it might take me to put this together and whether I will complete this at all. So far it is going well so staying optimistic.
", "rawMarkdown": "The data is still downloadable from the playground competition - one option would be to use it to train the model. On the other hand, it would be also interesting to verify there is no overlap between the train set from the previous comp and the test set here. Or if indeed all the images in the earlier train set are contained in this train set. Such analysis could be very useful.\n\nAs for bounding boxes, I am currently in the process of creating my own annotations. Aiming for around a 1000, maybe a bit more, we'll see. Planning on sharing them but it is not clear how long it might take me to put this together and whether I will complete this at all. So far it is going well so staying optimistic.", "votes": null }, { "id": "442660", "postDate": "12/20/2018 09:42:27", "content": "Thanks radek,\nBut it seems that https://www.kaggle.com/c/whale-categorization-playground/data has now exactly the same dataset as the current competition. So bounding boxes used in older kernels do not work.\n
@radek1 -- Would you mind sharing what tooling you were using for annotation? I'm thinking about starting up on this myself and have some time because of the holiday...
", "rawMarkdown": "radek1 -- Would you mind sharing what tooling you were using for annotation? I'm thinking about starting up on this myself and have some time because of the holiday...", "votes": null }, { "id": "445016", "postDate": "12/25/2018 11:16:51", "content": "I use sloth.
", "rawMarkdown": "I use [sloth](https://github.com/cvhciKIT/sloth).", "votes": null }, { "id": "445187", "postDate": "12/25/2018 21:12:26", "content": "Thanks, never used sloth. I'll give it a shot.
", "rawMarkdown": "Thanks, never used sloth. I'll give it a shot.", "votes": null }, { "id": "446321", "postDate": "12/27/2018 21:46:11", "content": "Thanks, appreciated.
", "rawMarkdown": "Thanks, appreciated.", "votes": null } ], "comments": [ { "id": 442623, "author_name": "radek1", "author_url": "", "post_date": "12/20/2018 08:21:18", "content": "The data is still downloadable from the playground competition - one option would be to use it to train the model. On the other hand, it would be also interesting to verify there is no overlap between the train set from the previous comp and the test set here. Or if indeed all the images in the earlier train set are contained in this train set. Such analysis could be very useful.
\n\nAs for bounding boxes, I am currently in the process of creating my own annotations. Aiming for around a 1000, maybe a bit more, we'll see. Planning on sharing them but it is not clear how long it might take me to put this together and whether I will complete this at all. So far it is going well so staying optimistic.
", "votes": null, "replies": [ { "id": 444781, "author_name": "larcat", "author_url": "", "post_date": "12/24/2018 19:40:09", "content": "@radek1 -- Would you mind sharing what tooling you were using for annotation? I'm thinking about starting up on this myself and have some time because of the holiday...
", "votes": null, "replies": [] }, { "id": 445016, "author_name": "radek1", "author_url": "", "post_date": "12/25/2018 11:16:51", "content": "I use sloth.
", "votes": null, "replies": [] }, { "id": 445187, "author_name": "badtyprr", "author_url": "", "post_date": "12/25/2018 21:12:26", "content": "Thanks, never used sloth. I'll give it a shot.
", "votes": null, "replies": [] }, { "id": 446321, "author_name": "larcat", "author_url": "", "post_date": "12/27/2018 21:46:11", "content": "Thanks, appreciated.
", "votes": null, "replies": [] } ] }, { "id": 442660, "author_name": "sakvaua", "author_url": "", "post_date": "12/20/2018 09:42:27", "content": "Thanks radek,\nBut it seems that https://www.kaggle.com/c/whale-categorization-playground/data has now exactly the same dataset as the current competition. So bounding boxes used in older kernels do not work.\n