Having the coordinates of the sea lions in an easy to process format (e.g. CSV with x,y,class) will be very useful for the purpose of counting, and can lead to more powerful solutions that can also output coordinates of the sea lions.
\n\nI spent some time trying to get the coordinates of sea lions using the dotted images, but, at least for me, this isn't going too great. I'm aiming for coordinates that are as good as the dots but my approaches so far are quite far from perfect. I might be wrong here, but given the image compression I think, generally, the process of extracting the coordinates from the dotted images will be noisy, generating suboptimal labels.
\n\nI'm not sure how the training dotted images were generated, but if they are automatically generated from stored coordinates labels, or if the coordinates labels are somehow held by NOAA, is there any chance we can have access to them as part of this competition?
\n\nThank you!
", "messages": [ { "id": "171836", "postDate": "03/31/2017 16:29:51", "content": "Having the coordinates of the sea lions in an easy to process format (e.g. CSV with x,y,class) will be very useful for the purpose of counting, and can lead to more powerful solutions that can also output coordinates of the sea lions.
\n\nI spent some time trying to get the coordinates of sea lions using the dotted images, but, at least for me, this isn't going too great. I'm aiming for coordinates that are as good as the dots but my approaches so far are quite far from perfect. I might be wrong here, but given the image compression I think, generally, the process of extracting the coordinates from the dotted images will be noisy, generating suboptimal labels.
\n\nI'm not sure how the training dotted images were generated, but if they are automatically generated from stored coordinates labels, or if the coordinates labels are somehow held by NOAA, is there any chance we can have access to them as part of this competition?
\n\nThank you!
", "rawMarkdown": "Having the coordinates of the sea lions in an easy to process format (e.g. CSV with x,y,class) will be very useful for the purpose of counting, and can lead to more powerful solutions that can also output coordinates of the sea lions.\n\nI spent some time trying to get the coordinates of sea lions using the dotted images, but, at least for me, this isn't going too great. I'm aiming for coordinates that are as good as the dots but my approaches so far are quite far from perfect. I might be wrong here, but given the image compression I think, generally, the process of extracting the coordinates from the dotted images will be noisy, generating suboptimal labels.\n\nI'm not sure how the training dotted images were generated, but if they are automatically generated from stored coordinates labels, or if the coordinates labels are somehow held by NOAA, is there any chance we can have access to them as part of this competition?\n\nThank you!", "votes": null }, { "id": "171856", "postDate": "03/31/2017 18:08:40", "content": "I have a prototype script that extracts coordinates. It's far from perfect. I've only run it on the small training sets so far, and it misses some edge cases.
\n\n", "rawMarkdown": "I have a prototype script that extracts coordinates. It's far from perfect. I've only run it on the small training sets so far, and it misses some edge cases. \n\nhttps://www.kaggle.com/threeplusone/noaa-fisheries-steller-sea-lion-population-count/sea-lion-coordinates", "votes": null }, { "id": "171860", "postDate": "03/31/2017 18:17:59", "content": "That's great! Very nice implementation! I will try it out on Monday, I guess it does better than what I've got so far. Thank you for sharing!
", "rawMarkdown": "That's great! Very nice implementation! I will try it out on Monday, I guess it does better than what I've got so far. Thank you for sharing!", "votes": null } ], "comments": [ { "id": 171856, "author_name": "threeplusone", "author_url": "", "post_date": "03/31/2017 18:08:40", "content": "I have a prototype script that extracts coordinates. It's far from perfect. I've only run it on the small training sets so far, and it misses some edge cases.
\n\n", "votes": null, "replies": [ { "id": 171860, "author_name": "velici", "author_url": "", "post_date": "03/31/2017 18:17:59", "content": "That's great! Very nice implementation! I will try it out on Monday, I guess it does better than what I've got so far. Thank you for sharing!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "171836": "Having the coordinates of the sea lions in an easy to process format (e.g. CSV with x,y,class) will be very useful for the purpose of counting, and can lead to more powerful solutions that can also output coordinates of the sea lions.\n\nI spent some time trying to get the coordinates of sea lions using the dotted images, but, at least for me, this isn't going too great. I'm aiming for coordinates that are as good as the dots but my approaches so far are quite far from perfect. I might be wrong here, but given the image compression I think, generally, the process of extracting the coordinates from the dotted images will be noisy, generating suboptimal labels.\n\nI'm not sure how the training dotted images were generated, but if they are automatically generated from stored coordinates labels, or if the coordinates labels are somehow held by NOAA, is there any chance we can have access to them as part of this competition?\n\nThank you!", "171856": "I have a prototype script that extracts coordinates. It's far from perfect. I've only run it on the small training sets so far, and it misses some edge cases. \n\nhttps://www.kaggle.com/threeplusone/noaa-fisheries-steller-sea-lion-population-count/sea-lion-coordinates", "171860": "That's great! Very nice implementation! I will try it out on Monday, I guess it does better than what I've got so far. Thank you for sharing!" }, "source": "meta" }