{ "id": 308278, "title": "Camera viewpoint dataset", "url": "/competitions/happy-whale-and-dolphin/discussion/308278", "author_name": "", "post_date": "2022-02-18T00:51:27.419310800Z", "votes": 12, "comment_count": 4, "views": 0, "content": "

Hi everyone!

\n

Following a comment by @lobrien, I got inspired and started to look into ways of detecting the camera viewpoints. Indeed, it could turn out to be an interesting tool to tell which side of an individual animal we are looking at.

\n

In the end, I spent a little over 3 hours labelling 1,680 images to tell whether the camera was the port or starboard side of the animal, or if it was unclear. I cannot guarantee that I did not make any mistakes as it could get a little tricky to tell for whales… Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!

\n

\"results-5-0\"

\n

I'd be keen to hear your opinion on how you would use such a dataset. If it is well received and helpful, I might annotate more data in the future!

\n

Here are the links to the dataset and a simple demo notebook.

", "messages": [ { "id": "1695114", "postDate": "02/18/2022 00:51:27", "content": "

Hi everyone!

\n

Following a comment by @lobrien, I got inspired and started to look into ways of detecting the camera viewpoints. Indeed, it could turn out to be an interesting tool to tell which side of an individual animal we are looking at.

\n

In the end, I spent a little over 3 hours labelling 1,680 images to tell whether the camera was the port or starboard side of the animal, or if it was unclear. I cannot guarantee that I did not make any mistakes as it could get a little tricky to tell for whales… Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!

\n

\"results-5-0\"

\n

I'd be keen to hear your opinion on how you would use such a dataset. If it is well received and helpful, I might annotate more data in the future!

\n

Here are the links to the dataset and a simple demo notebook.

", "rawMarkdown": "Hi everyone!\n\nFollowing a [comment](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoint-classification) by @lobrien, I got inspired and started to look into ways of detecting the camera viewpoints. Indeed, it could turn out to be an interesting tool to tell which side of an individual animal we are looking at.\n\nIn the end, I spent a little over 3 hours labelling 1,680 images to tell whether the camera was the port or starboard side of the animal, or if it was unclear. I cannot guarantee that I did not make any mistakes as it could get a little tricky to tell for whales... Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!\n\n\"results-5-0\"\n\n\nI'd be keen to hear your opinion on how you would use such a dataset. If it is well received and helpful, I might annotate more data in the future!\n\nHere are the links to the [dataset](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoints) and a simple [demo notebook](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoint-classification).", "votes": null }, { "id": "1695223", "postDate": "02/18/2022 03:00:15", "content": "

From your post:

\n
\n

Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!

\n
\n

This made me smile! 🙏

\n

Thanks for creating the dataset! @frlemarchand, may I ask what tool did you use for labeling?

", "rawMarkdown": "From your post:\n> Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!\n\nThis made me smile! 🙏\n\nThanks for creating the dataset! @frlemarchand, may I ask what tool did you use for labeling?", "votes": null }, { "id": "1695798", "postDate": "02/18/2022 11:28:03", "content": "

I looked into different tools but did not find anything simple enough so I've quickly written my own tool. I loop through the rows of my dataframe and display the original image, as well as a cropped version of the image, and give an input. That's it!

", "rawMarkdown": "I looked into different tools but did not find anything simple enough so I've quickly written my own tool. I loop through the rows of my dataframe and display the original image, as well as a [cropped version of the image](https://www.kaggle.com/phalanx/whale2-cropped-dataset), and give an input. That's it!", "votes": null }, { "id": "1695862", "postDate": "02/18/2022 12:12:03", "content": "

Thanks for the reply!

\n

There was actually a lecture in fastai that showcased how to use Jupyter widgets to do the same too, I was curious if your approach was similar-sounds like so indeed. :D

", "rawMarkdown": "Thanks for the reply!\n\nThere was actually a lecture in fastai that showcased how to use Jupyter widgets to do the same too, I was curious if your approach was similar-sounds like so indeed. :D", "votes": null }, { "id": "1718709", "postDate": "03/11/2022 05:09:48", "content": "

Would splitting the dataset into directional views be helpful for training? ie. This is what a Dolphin looks like from the front… and this is what it looks like from the side? Then use different models depending on the confidence and/or outcome?

", "rawMarkdown": "Would splitting the dataset into directional views be helpful for training? ie. This is what a Dolphin looks like from the front... and this is what it looks like from the side? Then use different models depending on the confidence and/or outcome?", "votes": null } ], "comments": [ { "id": 1695223, "author_name": "init27", "author_url": "", "post_date": "02/18/2022 03:00:15", "content": "

From your post:

\n
\n

Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!

\n
\n

This made me smile! 🙏

\n

Thanks for creating the dataset! @frlemarchand, may I ask what tool did you use for labeling?

", "votes": null, "replies": [ { "id": 1695798, "author_name": "frlemarchand", "author_url": "", "post_date": "02/18/2022 11:28:03", "content": "

I looked into different tools but did not find anything simple enough so I've quickly written my own tool. I loop through the rows of my dataframe and display the original image, as well as a cropped version of the image, and give an input. That's it!

", "votes": null, "replies": [] }, { "id": 1695862, "author_name": "init27", "author_url": "", "post_date": "02/18/2022 12:12:03", "content": "

Thanks for the reply!

\n

There was actually a lecture in fastai that showcased how to use Jupyter widgets to do the same too, I was curious if your approach was similar-sounds like so indeed. :D

", "votes": null, "replies": [] } ] }, { "id": 1718709, "author_name": "dentistdad", "author_url": "", "post_date": "03/11/2022 05:09:48", "content": "

Would splitting the dataset into directional views be helpful for training? ie. This is what a Dolphin looks like from the front… and this is what it looks like from the side? Then use different models depending on the confidence and/or outcome?

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1695114": "Hi everyone!\n\nFollowing a [comment](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoint-classification) by @lobrien, I got inspired and started to look into ways of detecting the camera viewpoints. Indeed, it could turn out to be an interesting tool to tell which side of an individual animal we are looking at.\n\nIn the end, I spent a little over 3 hours labelling 1,680 images to tell whether the camera was the port or starboard side of the animal, or if it was unclear. I cannot guarantee that I did not make any mistakes as it could get a little tricky to tell for whales... Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!\n\n\"results-5-0\"\n\n\nI'd be keen to hear your opinion on how you would use such a dataset. If it is well received and helpful, I might annotate more data in the future!\n\nHere are the links to the [dataset](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoints) and a simple [demo notebook](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoint-classification).", "1695223": "From your post:\n> Beluga whales can really look like giant mozzarellas floating in the middle of the ocean!\n\nThis made me smile! 🙏\n\nThanks for creating the dataset! @frlemarchand, may I ask what tool did you use for labeling?", "1695798": "I looked into different tools but did not find anything simple enough so I've quickly written my own tool. I loop through the rows of my dataframe and display the original image, as well as a [cropped version of the image](https://www.kaggle.com/phalanx/whale2-cropped-dataset), and give an input. That's it!", "1695862": "Thanks for the reply!\n\nThere was actually a lecture in fastai that showcased how to use Jupyter widgets to do the same too, I was curious if your approach was similar-sounds like so indeed. :D", "1718709": "Would splitting the dataset into directional views be helpful for training? ie. This is what a Dolphin looks like from the front... and this is what it looks like from the side? Then use different models depending on the confidence and/or outcome?" }, "source": "meta" }