{ "id": 21037, "title": "Good job Vladimir", "url": "/competitions/draper-satellite-image-chronology/discussion/21037", "author_name": "", "post_date": "2016-05-18T04:11:17.443Z", "votes": null, "comment_count": 15, "views": 2355, "content": "

1.00000

", "messages": [ { "id": "120423", "postDate": "05/18/2016 04:11:17", "content": "

1.00000

", "rawMarkdown": "1.00000", "votes": null }, { "id": "120424", "postDate": "05/18/2016 05:02:33", "content": "

When this competition first started I thought it might be fun to try a completely manual approach and see how that stacked up. However it gave me a headache and was way more boring than I thought it would be so I quickly abandoned it. I am really glad now that I didn't waste my time on that approach ;).

", "rawMarkdown": "When this competition first started I thought it might be fun to try a completely manual approach and see how that stacked up. However it gave me a headache and was way more boring than I thought it would be so I quickly abandoned it. I am really glad now that I didn't waste my time on that approach ;).", "votes": null }, { "id": "120434", "postDate": "05/18/2016 07:01:26", "content": "

Congratulations Vladimir. I'm still completely stumped as to how you've done this (which is driving me a little insane!).

\n\n

It will be interesting to see if you make any further submissions as the competition progresses.

", "rawMarkdown": "Congratulations Vladimir. I'm still completely stumped as to how you've done this (which is driving me a little insane!). \r\n\r\nIt will be interesting to see if you make any further submissions as the competition progresses.", "votes": null }, { "id": "120436", "postDate": "05/18/2016 07:31:59", "content": "

Congrats Vladimir. In 2 submissions. Amazing

", "rawMarkdown": "Congrats Vladimir. In 2 submissions. Amazing", "votes": null }, { "id": "120438", "postDate": "05/18/2016 07:38:26", "content": "

either overfitting or a leak :)

", "rawMarkdown": "either overfitting or a leak :)", "votes": null }, { "id": "120443", "postDate": "05/18/2016 08:49:43", "content": "

Actually if you are sure that you, @Vladimir , have a score of 1 in the private set you can tell us how you did it because we can't beat you anymore ;)

", "rawMarkdown": "Actually if you are sure that you, @Vladimir , have a score of 1 in the private set you can tell us how you did it because we can't beat you anymore ;)", "votes": null }, { "id": "120445", "postDate": "05/18/2016 09:48:25", "content": "

Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\nThough I think some other steps can be automated, I preferred to do it by hand, because I'm pretty new to OpenCV and I was aware of the fact, that the perfect score can be achieved and I wanted to be the first.

\n\n

Now I have a lot of time to automate, but some work will be still left for a human. I can't imagine how this could be fully automated.

\n\n

I'm still not sure if I have it all right. Some sets were really challenging - like set162.

", "rawMarkdown": "Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\r\nThough I think some other steps can be automated, I preferred to do it by hand, because I'm pretty new to OpenCV and I was aware of the fact, that the perfect score can be achieved and I wanted to be the first.\r\n\r\nNow I have a lot of time to automate, but some work will be still left for a human. I can't imagine how this could be fully automated.\r\n\r\nI'm still not sure if I have it all right. Some sets were really challenging - like set162.", "votes": null }, { "id": "120454", "postDate": "05/18/2016 11:27:26", "content": "

[quote=Abhishek;120438]

\n\n

either overfitting or a leak :)

\n\n

[/quote]\nSince this isn't a ML problem, overfitting doesn't seem very relevant. I know some people are trying to solve it that way but obviously they won't achieve perfect solutions.

\n\n

[quote=Vladimir Tomecek;120445]

\n\n

I'm still not sure if I have it all right. Some sets were really challenging - like set162.

\n\n

[/quote]\nTell me about it. The images with almost nothing in them are such a pain to register correctly.

", "rawMarkdown": "[quote=Abhishek;120438]\r\n\r\neither overfitting or a leak :)\r\n\r\n[/quote]\r\nSince this isn't a ML problem, overfitting doesn't seem very relevant. I know some people are trying to solve it that way but obviously they won't achieve perfect solutions.\r\n\r\n[quote=Vladimir Tomecek;120445]\r\n\r\nI'm still not sure if I have it all right. Some sets were really challenging - like set162.\r\n\r\n[/quote]\r\nTell me about it. The images with almost nothing in them are such a pain to register correctly.", "votes": null }, { "id": "120456", "postDate": "05/18/2016 12:15:39", "content": "

[quote=Vladimir Tomecek;120445]

\n\n

Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.

\n\n

[/quote]

\n\n

Thanks for sharing. I assumed that you'd found some clever ML approach or leakage, rather than hand labeling. Having stared at Googlemaps for far too long, I'm impressed that you've done this so quickly and within 2 submissions. Well done!

", "rawMarkdown": "[quote=Vladimir Tomecek;120445]\r\n\r\nThanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\r\n\r\n[/quote]\r\n\r\nThanks for sharing. I assumed that you'd found some clever ML approach or leakage, rather than hand labeling. Having stared at Googlemaps for far too long, I'm impressed that you've done this so quickly and within 2 submissions. Well done!", "votes": null }, { "id": "120459", "postDate": "05/18/2016 12:46:15", "content": "

It seems like we will have more 1. in the next several days. Anw, I think reaching 1. in only 2 submissions is very impressive. Dont know others can reproduce a similar result in the 40-day left of the competition, i.e 1.0 in just one submission.

\n\n

Btw, what would happen if we have more than three perfect scores at the end? Did it happen before?

", "rawMarkdown": "It seems like we will have more 1. in the next several days. Anw, I think reaching 1. in only 2 submissions is very impressive. Dont know others can reproduce a similar result in the 40-day left of the competition, i.e 1.0 in just one submission.\r\n\r\nBtw, what would happen if we have more than three perfect scores at the end? Did it happen before?", "votes": null }, { "id": "120462", "postDate": "05/18/2016 13:25:20", "content": "

[quote=all_random;120459]

\n\n

Btw, what would happen if we have more than three perfect scores at the end? Did it happen before?

\n\n

[/quote]

\n\n

On kaggle, submission time was used to break the ties in score. Namely, for two teams which achieved exactly the same score on the private LB, the earlier that submission was made, the better the team would be ranked.

\n\n

Best regards,

\n\n

Shize

", "rawMarkdown": "[quote=all_random;120459]\r\n\r\nBtw, what would happen if we have more than three perfect scores at the end? Did it happen before?\r\n\r\n[/quote]\r\n\r\nOn kaggle, submission time was used to break the ties in score. Namely, for two teams which achieved exactly the same score on the private LB, the earlier that submission was made, the better the team would be ranked.\r\n\r\nBest regards,\r\n\r\nShize", "votes": null }, { "id": "120470", "postDate": "05/18/2016 14:33:27", "content": "

Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D

", "rawMarkdown": "Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\r\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D", "votes": null }, { "id": "120471", "postDate": "05/18/2016 14:40:04", "content": "

[quote=woshialex;120470]

\n\n

Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D

\n\n

[/quote]\nThe public LB is only comparing against less than 17% of the sets. Since both the leaders right now have had lower scores, presumably their method isn't foolproof. I'd be willing to bet that there are a couple of sets that are dastardly hard to get right that end up tripping up some of the 1.0000s on the final leader-board. So there's still an incentive to continue through June if you think you can hit perfect.

", "rawMarkdown": "[quote=woshialex;120470]\r\n\r\nReally? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\r\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D\r\n\r\n[/quote]\r\nThe public LB is only comparing against less than 17% of the sets. Since both the leaders right now have had lower scores, presumably their method isn't foolproof. I'd be willing to bet that there are a couple of sets that are dastardly hard to get right that end up tripping up some of the 1.0000s on the final leader-board. So there's still an incentive to continue through June if you think you can hit perfect.", "votes": null }, { "id": "120511", "postDate": "05/18/2016 19:04:26", "content": "

I've tried hand labelling the data - and even then I've struggled to work out the order. Congratulations!

", "rawMarkdown": "I've tried hand labelling the data - and even then I've struggled to work out the order. Congratulations!", "votes": null }, { "id": "120525", "postDate": "05/18/2016 21:14:28", "content": "

If this is not a leak or something weird in the data... it seems worth more than $40k for reliability.

", "rawMarkdown": "If this is not a leak or something weird in the data... it seems worth more than $40k for reliability.", "votes": null }, { "id": "120561", "postDate": "05/19/2016 03:38:54", "content": "

How do you hand label the images? I can't see any clue. They all look the same.

", "rawMarkdown": "How do you hand label the images? I can't see any clue. They all look the same.", "votes": null } ], "comments": [ { "id": 120424, "author_name": "jturkewitz", "author_url": "", "post_date": "05/18/2016 05:02:33", "content": "

When this competition first started I thought it might be fun to try a completely manual approach and see how that stacked up. However it gave me a headache and was way more boring than I thought it would be so I quickly abandoned it. I am really glad now that I didn't waste my time on that approach ;).

", "votes": null, "replies": [] }, { "id": 120434, "author_name": "nigelcarpenter", "author_url": "", "post_date": "05/18/2016 07:01:26", "content": "

Congratulations Vladimir. I'm still completely stumped as to how you've done this (which is driving me a little insane!).

\n\n

It will be interesting to see if you make any further submissions as the competition progresses.

", "votes": null, "replies": [] }, { "id": 120436, "author_name": "santiagomota", "author_url": "", "post_date": "05/18/2016 07:31:59", "content": "

Congrats Vladimir. In 2 submissions. Amazing

", "votes": null, "replies": [] }, { "id": 120438, "author_name": "abhishek", "author_url": "", "post_date": "05/18/2016 07:38:26", "content": "

either overfitting or a leak :)

", "votes": null, "replies": [] }, { "id": 120443, "author_name": "wikunia", "author_url": "", "post_date": "05/18/2016 08:49:43", "content": "

Actually if you are sure that you, @Vladimir , have a score of 1 in the private set you can tell us how you did it because we can't beat you anymore ;)

", "votes": null, "replies": [] }, { "id": 120445, "author_name": "vtomecek", "author_url": "", "post_date": "05/18/2016 09:48:25", "content": "

Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\nThough I think some other steps can be automated, I preferred to do it by hand, because I'm pretty new to OpenCV and I was aware of the fact, that the perfect score can be achieved and I wanted to be the first.

\n\n

Now I have a lot of time to automate, but some work will be still left for a human. I can't imagine how this could be fully automated.

\n\n

I'm still not sure if I have it all right. Some sets were really challenging - like set162.

", "votes": null, "replies": [] }, { "id": 120454, "author_name": "tylervigen", "author_url": "", "post_date": "05/18/2016 11:27:26", "content": "

[quote=Abhishek;120438]

\n\n

either overfitting or a leak :)

\n\n

[/quote]\nSince this isn't a ML problem, overfitting doesn't seem very relevant. I know some people are trying to solve it that way but obviously they won't achieve perfect solutions.

\n\n

[quote=Vladimir Tomecek;120445]

\n\n

I'm still not sure if I have it all right. Some sets were really challenging - like set162.

\n\n

[/quote]\nTell me about it. The images with almost nothing in them are such a pain to register correctly.

", "votes": null, "replies": [] }, { "id": 120456, "author_name": "nigelcarpenter", "author_url": "", "post_date": "05/18/2016 12:15:39", "content": "

[quote=Vladimir Tomecek;120445]

\n\n

Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.

\n\n

[/quote]

\n\n

Thanks for sharing. I assumed that you'd found some clever ML approach or leakage, rather than hand labeling. Having stared at Googlemaps for far too long, I'm impressed that you've done this so quickly and within 2 submissions. Well done!

", "votes": null, "replies": [] }, { "id": 120459, "author_name": "a11rand0m", "author_url": "", "post_date": "05/18/2016 12:46:15", "content": "

It seems like we will have more 1. in the next several days. Anw, I think reaching 1. in only 2 submissions is very impressive. Dont know others can reproduce a similar result in the 40-day left of the competition, i.e 1.0 in just one submission.

\n\n

Btw, what would happen if we have more than three perfect scores at the end? Did it happen before?

", "votes": null, "replies": [] }, { "id": 120462, "author_name": "sushize", "author_url": "", "post_date": "05/18/2016 13:25:20", "content": "

[quote=all_random;120459]

\n\n

Btw, what would happen if we have more than three perfect scores at the end? Did it happen before?

\n\n

[/quote]

\n\n

On kaggle, submission time was used to break the ties in score. Namely, for two teams which achieved exactly the same score on the private LB, the earlier that submission was made, the better the team would be ranked.

\n\n

Best regards,

\n\n

Shize

", "votes": null, "replies": [] }, { "id": 120470, "author_name": "woshialex", "author_url": "", "post_date": "05/18/2016 14:33:27", "content": "

Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D

", "votes": null, "replies": [] }, { "id": 120471, "author_name": "tylervigen", "author_url": "", "post_date": "05/18/2016 14:40:04", "content": "

[quote=woshialex;120470]

\n\n

Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D

\n\n

[/quote]\nThe public LB is only comparing against less than 17% of the sets. Since both the leaders right now have had lower scores, presumably their method isn't foolproof. I'd be willing to bet that there are a couple of sets that are dastardly hard to get right that end up tripping up some of the 1.0000s on the final leader-board. So there's still an incentive to continue through June if you think you can hit perfect.

", "votes": null, "replies": [] }, { "id": 120511, "author_name": "anokas", "author_url": "", "post_date": "05/18/2016 19:04:26", "content": "

I've tried hand labelling the data - and even then I've struggled to work out the order. Congratulations!

", "votes": null, "replies": [] }, { "id": 120525, "author_name": "carloshuertas", "author_url": "", "post_date": "05/18/2016 21:14:28", "content": "

If this is not a leak or something weird in the data... it seems worth more than $40k for reliability.

", "votes": null, "replies": [] }, { "id": 120561, "author_name": "ozgung", "author_url": "", "post_date": "05/19/2016 03:38:54", "content": "

How do you hand label the images? I can't see any clue. They all look the same.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "120423": "1.00000", "120424": "When this competition first started I thought it might be fun to try a completely manual approach and see how that stacked up. However it gave me a headache and was way more boring than I thought it would be so I quickly abandoned it. I am really glad now that I didn't waste my time on that approach ;).", "120434": "Congratulations Vladimir. I'm still completely stumped as to how you've done this (which is driving me a little insane!). \r\n\r\nIt will be interesting to see if you make any further submissions as the competition progresses.", "120436": "Congrats Vladimir. In 2 submissions. Amazing", "120438": "either overfitting or a leak :)", "120443": "Actually if you are sure that you, @Vladimir , have a score of 1 in the private set you can tell us how you did it because we can't beat you anymore ;)", "120445": "Thanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\r\nThough I think some other steps can be automated, I preferred to do it by hand, because I'm pretty new to OpenCV and I was aware of the fact, that the perfect score can be achieved and I wanted to be the first.\r\n\r\nNow I have a lot of time to automate, but some work will be still left for a human. I can't imagine how this could be fully automated.\r\n\r\nI'm still not sure if I have it all right. Some sets were really challenging - like set162.", "120454": "[quote=Abhishek;120438]\r\n\r\neither overfitting or a leak :)\r\n\r\n[/quote]\r\nSince this isn't a ML problem, overfitting doesn't seem very relevant. I know some people are trying to solve it that way but obviously they won't achieve perfect solutions.\r\n\r\n[quote=Vladimir Tomecek;120445]\r\n\r\nI'm still not sure if I have it all right. Some sets were really challenging - like set162.\r\n\r\n[/quote]\r\nTell me about it. The images with almost nothing in them are such a pain to register correctly.", "120456": "[quote=Vladimir Tomecek;120445]\r\n\r\nThanks, it was mostly manual work with the help of Google Maps and OpenCV's stitching functionality.\r\n\r\n[/quote]\r\n\r\nThanks for sharing. I assumed that you'd found some clever ML approach or leakage, rather than hand labeling. Having stared at Googlemaps for far too long, I'm impressed that you've done this so quickly and within 2 submissions. Well done!", "120459": "It seems like we will have more 1. in the next several days. Anw, I think reaching 1. in only 2 submissions is very impressive. Dont know others can reproduce a similar result in the 40-day left of the competition, i.e 1.0 in just one submission.\r\n\r\nBtw, what would happen if we have more than three perfect scores at the end? Did it happen before?", "120462": "[quote=all_random;120459]\r\n\r\nBtw, what would happen if we have more than three perfect scores at the end? Did it happen before?\r\n\r\n[/quote]\r\n\r\nOn kaggle, submission time was used to break the ties in score. Namely, for two teams which achieved exactly the same score on the private LB, the earlier that submission was made, the better the team would be ranked.\r\n\r\nBest regards,\r\n\r\nShize", "120470": "Really? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\r\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D", "120471": "[quote=woshialex;120470]\r\n\r\nReally? So if we have 10 perfect 1.0 at the end, only the first three that achieve 1.0 can win the prize?\r\nThat means once we have 3 1.0s there is not much attraction for the rest of us, people need to move fast :D\r\n\r\n[/quote]\r\nThe public LB is only comparing against less than 17% of the sets. Since both the leaders right now have had lower scores, presumably their method isn't foolproof. I'd be willing to bet that there are a couple of sets that are dastardly hard to get right that end up tripping up some of the 1.0000s on the final leader-board. So there's still an incentive to continue through June if you think you can hit perfect.", "120511": "I've tried hand labelling the data - and even then I've struggled to work out the order. Congratulations!", "120525": "If this is not a leak or something weird in the data... it seems worth more than $40k for reliability.", "120561": "How do you hand label the images? I can't see any clue. They all look the same." }, "source": "meta" }