{ "id": 187725, "title": "Congratulations!", "url": "/competitions/landmark-recognition-2020/discussion/187725", "author_name": "", "post_date": "2020-09-30T03:55:57.465723500Z", "votes": 33, "comment_count": 19, "views": 0, "content": "

Congratulations to @philippsinger @christofhenkel @bestfitting and all other winners! You guys have done a great job on this harsh competition!

\n
\n

We worked very hard on global features while unfortunately we failed to train local features well in our models.

\n

So we abandoned local features in our final solution and use global features only. Fortunately we stayed in top3 finally.

\n

Our brief summary of solution is released: https://www.kaggle.com/c/landmark-recognition-2020/discussion/187757

\n

Congratulations again!

", "messages": [ { "id": "1032233", "postDate": "09/30/2020 03:55:57", "content": "

Congratulations to @philippsinger @christofhenkel @bestfitting and all other winners! You guys have done a great job on this harsh competition!

\n
\n

We worked very hard on global features while unfortunately we failed to train local features well in our models.

\n

So we abandoned local features in our final solution and use global features only. Fortunately we stayed in top3 finally.

\n

Our brief summary of solution is released: https://www.kaggle.com/c/landmark-recognition-2020/discussion/187757

\n

Congratulations again!

", "rawMarkdown": "Congratulations to @philippsinger @christofhenkel @bestfitting and all other winners! You guys have done a great job on this harsh competition!\n\n---\n\nWe worked very hard on global features while unfortunately we failed to train local features well in our models.\n\nSo we abandoned local features in our final solution and use global features only. Fortunately we stayed in top3 finally.\n\nOur brief summary of solution is released: https://www.kaggle.com/c/landmark-recognition-2020/discussion/187757\n\nCongratulations again!", "votes": null }, { "id": "1032321", "postDate": "09/30/2020 05:54:00", "content": "

Great, Congratulations 😍
\nWaiting for the summary of your solution.

", "rawMarkdown": "Great, Congratulations 😍\nWaiting for the summary of your solution.", "votes": null }, { "id": "1032388", "postDate": "09/30/2020 07:08:53", "content": "

Congrats to your team. Using global features only to get 3rd place was very impressive. I'm eager to learn more from your team's solution 😄

", "rawMarkdown": "Congrats to your team. Using global features only to get 3rd place was very impressive. I'm eager to learn more from your team's solution 😄", "votes": null }, { "id": "1032425", "postDate": "09/30/2020 07:38:52", "content": "

Thanks and congrats to you as well! Our solution also uses global features only. We tried hard to get local features to work, but on the last day they either gave score 0 or timed out. That said, on CV they only gave minimal boosts, so I would not expect that they would have contributed too much.

", "rawMarkdown": "Thanks and congrats to you as well! Our solution also uses global features only. We tried hard to get local features to work, but on the last day they either gave score 0 or timed out. That said, on CV they only gave minimal boosts, so I would not expect that they would have contributed too much.", "votes": null }, { "id": "1032449", "postDate": "09/30/2020 07:59:51", "content": "

Congratulations to all and especially the winners! I'm really impressed by the winning margin the top 3 have over the rest of the leaderboard. I'm sure a great deal of work has been done and most likely some useful techniques discovered!

", "rawMarkdown": "Congratulations to all and especially the winners! I'm really impressed by the winning margin the top 3 have over the rest of the leaderboard. I'm sure a great deal of work has been done and most likely some useful techniques discovered!", "votes": null }, { "id": "1032524", "postDate": "09/30/2020 09:06:38", "content": "

wow, It's impressived that your team also use global features only, but got a wonderful public/private score.

", "rawMarkdown": "wow, It's impressived that your team also use global features only, but got a wonderful public/private score.", "votes": null }, { "id": "1032621", "postDate": "09/30/2020 10:16:08", "content": "

Great job! Did you guys really used cycleGAN?
\nCan't wait for reading your writeup!

", "rawMarkdown": "Great job! Did you guys really used cycleGAN?\nCan't wait for reading your writeup!", "votes": null }, { "id": "1032646", "postDate": "09/30/2020 10:37:11", "content": "

Wow, another top team with global descriptors only. So, you guys used CycleGANs for augmentation? I'm waiting for your writeup with excitement!

", "rawMarkdown": "Wow, another top team with global descriptors only. So, you guys used CycleGANs for augmentation? I'm waiting for your writeup with excitement!", "votes": null }, { "id": "1032650", "postDate": "09/30/2020 10:40:56", "content": "

Congratulations to all the winners!

", "rawMarkdown": "Congratulations to all the winners!", "votes": null }, { "id": "1032667", "postDate": "09/30/2020 10:59:57", "content": "

No, unfortunately we did not use CycleGAN in the end. We used it in SIIM for some experiments to align images and thought that maybe if nothing else will pop up we might try it so we put it in the list as backup. After all, it was used in some image comps lately with success (e.g., Bengali). Actually, in the end we only need timm package and gldv2 data.

", "rawMarkdown": "No, unfortunately we did not use CycleGAN in the end. We used it in SIIM for some experiments to align images and thought that maybe if nothing else will pop up we might try it so we put it in the list as backup. After all, it was used in some image comps lately with success (e.g., Bengali). Actually, in the end we only need timm package and gldv2 data.", "votes": null }, { "id": "1032705", "postDate": "09/30/2020 11:44:48", "content": "

Want to read your writeup!

", "rawMarkdown": "Want to read your writeup!", "votes": null }, { "id": "1033219", "postDate": "09/30/2020 18:40:27", "content": "

Congratulations! This is my first comment ;-)

", "rawMarkdown": "Congratulations! This is my first comment ;-)", "votes": null }, { "id": "1033714", "postDate": "10/01/2020 08:01:36", "content": "

wow that's cool

", "rawMarkdown": "wow that's cool", "votes": null }, { "id": "1034130", "postDate": "10/01/2020 14:24:01", "content": "

Congratulations :-)

", "rawMarkdown": "Congratulations :-)", "votes": null }, { "id": "1035337", "postDate": "10/02/2020 16:40:37", "content": "

@philippsinger did you guys had many score 0 in the last competition day? We had 9 zero scores in our last 9 subs. We spend some time to check the subs and submitted also notebooks that worked before with no success.

", "rawMarkdown": "philippsinger did you guys had many score 0 in the last competition day? We had 9 zero scores in our last 9 subs. We spend some time to check the subs and submitted also notebooks that worked before with no success.", "votes": null }, { "id": "1035353", "postDate": "10/02/2020 16:50:32", "content": "

Congratulations!!

", "rawMarkdown": "Congratulations!!", "votes": null }, { "id": "1035355", "postDate": "10/02/2020 16:53:05", "content": "

@Giba we had two of them. But it was 99% a fault on our side with GPU memory running out and the final sub just being the constant one. After deleting models and reducing batch size etc, it seemed to work, but had not time to finish it. So maybe if you added more models, or other things, this might have been the issue as well.

", "rawMarkdown": "Giba we had two of them. But it was 99% a fault on our side with GPU memory running out and the final sub just being the constant one. After deleting models and reducing batch size etc, it seemed to work, but had not time to finish it. So maybe if you added more models, or other things, this might have been the issue as well.", "votes": null }, { "id": "1035534", "postDate": "10/02/2020 19:13:44", "content": "

Congratulations
\nAwesome work Thanks for sharing your solution overview with us

", "rawMarkdown": "Congratulations \nAwesome work Thanks for sharing your solution overview with us", "votes": null }, { "id": "1037249", "postDate": "10/04/2020 18:21:45", "content": "

Congratulations to all the hard working people.

", "rawMarkdown": "Congratulations to all the hard working people.", "votes": null }, { "id": "1493221", "postDate": "08/27/2021 17:48:09", "content": "

congratulations!

", "rawMarkdown": "congratulations!", "votes": null } ], "comments": [ { "id": 1032321, "author_name": "rahim3", "author_url": "", "post_date": "09/30/2020 05:54:00", "content": "

Great, Congratulations 😍
\nWaiting for the summary of your solution.

", "votes": null, "replies": [] }, { "id": 1032388, "author_name": "andy2709", "author_url": "", "post_date": "09/30/2020 07:08:53", "content": "

Congrats to your team. Using global features only to get 3rd place was very impressive. I'm eager to learn more from your team's solution 😄

", "votes": null, "replies": [] }, { "id": 1032425, "author_name": "philippsinger", "author_url": "", "post_date": "09/30/2020 07:38:52", "content": "

Thanks and congrats to you as well! Our solution also uses global features only. We tried hard to get local features to work, but on the last day they either gave score 0 or timed out. That said, on CV they only gave minimal boosts, so I would not expect that they would have contributed too much.

", "votes": null, "replies": [ { "id": 1032524, "author_name": "garybios", "author_url": "", "post_date": "09/30/2020 09:06:38", "content": "

wow, It's impressived that your team also use global features only, but got a wonderful public/private score.

", "votes": null, "replies": [] }, { "id": 1032621, "author_name": "haqishen", "author_url": "", "post_date": "09/30/2020 10:16:08", "content": "

Great job! Did you guys really used cycleGAN?
\nCan't wait for reading your writeup!

", "votes": null, "replies": [] }, { "id": 1032646, "author_name": "chankhavu", "author_url": "", "post_date": "09/30/2020 10:37:11", "content": "

Wow, another top team with global descriptors only. So, you guys used CycleGANs for augmentation? I'm waiting for your writeup with excitement!

", "votes": null, "replies": [] }, { "id": 1032667, "author_name": "philippsinger", "author_url": "", "post_date": "09/30/2020 10:59:57", "content": "

No, unfortunately we did not use CycleGAN in the end. We used it in SIIM for some experiments to align images and thought that maybe if nothing else will pop up we might try it so we put it in the list as backup. After all, it was used in some image comps lately with success (e.g., Bengali). Actually, in the end we only need timm package and gldv2 data.

", "votes": null, "replies": [] }, { "id": 1035337, "author_name": "titericz", "author_url": "", "post_date": "10/02/2020 16:40:37", "content": "

@philippsinger did you guys had many score 0 in the last competition day? We had 9 zero scores in our last 9 subs. We spend some time to check the subs and submitted also notebooks that worked before with no success.

", "votes": null, "replies": [] }, { "id": 1035355, "author_name": "philippsinger", "author_url": "", "post_date": "10/02/2020 16:53:05", "content": "

@Giba we had two of them. But it was 99% a fault on our side with GPU memory running out and the final sub just being the constant one. After deleting models and reducing batch size etc, it seemed to work, but had not time to finish it. So maybe if you added more models, or other things, this might have been the issue as well.

", "votes": null, "replies": [] } ] }, { "id": 1032449, "author_name": "andypenrose", "author_url": "", "post_date": "09/30/2020 07:59:51", "content": "

Congratulations to all and especially the winners! I'm really impressed by the winning margin the top 3 have over the rest of the leaderboard. I'm sure a great deal of work has been done and most likely some useful techniques discovered!

", "votes": null, "replies": [] }, { "id": 1032650, "author_name": "isiomaidemudia", "author_url": "", "post_date": "09/30/2020 10:40:56", "content": "

Congratulations to all the winners!

", "votes": null, "replies": [] }, { "id": 1032705, "author_name": "dsamitgarg", "author_url": "", "post_date": "09/30/2020 11:44:48", "content": "

Want to read your writeup!

", "votes": null, "replies": [] }, { "id": 1033219, "author_name": "juanfranmartn", "author_url": "", "post_date": "09/30/2020 18:40:27", "content": "

Congratulations! This is my first comment ;-)

", "votes": null, "replies": [] }, { "id": 1033714, "author_name": "mhoali", "author_url": "", "post_date": "10/01/2020 08:01:36", "content": "

wow that's cool

", "votes": null, "replies": [] }, { "id": 1034130, "author_name": "alistairsedwell", "author_url": "", "post_date": "10/01/2020 14:24:01", "content": "

Congratulations :-)

", "votes": null, "replies": [] }, { "id": 1035353, "author_name": "rahuldshetty", "author_url": "", "post_date": "10/02/2020 16:50:32", "content": "

Congratulations!!

", "votes": null, "replies": [] }, { "id": 1035534, "author_name": "surekharamireddy", "author_url": "", "post_date": "10/02/2020 19:13:44", "content": "

Congratulations
\nAwesome work Thanks for sharing your solution overview with us

", "votes": null, "replies": [] }, { "id": 1493221, "author_name": "alexiadeboynes", "author_url": "", "post_date": "08/27/2021 17:48:09", "content": "

congratulations!

", "votes": null, "replies": [] }, { "id": 1037249, "author_name": "alincijov", "author_url": "", "post_date": "10/04/2020 18:21:45", "content": "

Congratulations to all the hard working people.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1032233": "Congratulations to @philippsinger @christofhenkel @bestfitting and all other winners! You guys have done a great job on this harsh competition!\n\n---\n\nWe worked very hard on global features while unfortunately we failed to train local features well in our models.\n\nSo we abandoned local features in our final solution and use global features only. Fortunately we stayed in top3 finally.\n\nOur brief summary of solution is released: https://www.kaggle.com/c/landmark-recognition-2020/discussion/187757\n\nCongratulations again!", "1032321": "Great, Congratulations 😍\nWaiting for the summary of your solution.", "1032388": "Congrats to your team. Using global features only to get 3rd place was very impressive. I'm eager to learn more from your team's solution 😄", "1032425": "Thanks and congrats to you as well! Our solution also uses global features only. We tried hard to get local features to work, but on the last day they either gave score 0 or timed out. That said, on CV they only gave minimal boosts, so I would not expect that they would have contributed too much.", "1032449": "Congratulations to all and especially the winners! I'm really impressed by the winning margin the top 3 have over the rest of the leaderboard. I'm sure a great deal of work has been done and most likely some useful techniques discovered!", "1032524": "wow, It's impressived that your team also use global features only, but got a wonderful public/private score.", "1032621": "Great job! Did you guys really used cycleGAN?\nCan't wait for reading your writeup!", "1032646": "Wow, another top team with global descriptors only. So, you guys used CycleGANs for augmentation? I'm waiting for your writeup with excitement!", "1032650": "Congratulations to all the winners!", "1032667": "No, unfortunately we did not use CycleGAN in the end. We used it in SIIM for some experiments to align images and thought that maybe if nothing else will pop up we might try it so we put it in the list as backup. After all, it was used in some image comps lately with success (e.g., Bengali). Actually, in the end we only need timm package and gldv2 data.", "1032705": "Want to read your writeup!", "1033219": "Congratulations! This is my first comment ;-)", "1033714": "wow that's cool", "1034130": "Congratulations :-)", "1035337": "philippsinger did you guys had many score 0 in the last competition day? We had 9 zero scores in our last 9 subs. We spend some time to check the subs and submitted also notebooks that worked before with no success.", "1035353": "Congratulations!!", "1035355": "Giba we had two of them. But it was 99% a fault on our side with GPU memory running out and the final sub just being the constant one. After deleting models and reducing batch size etc, it seemed to work, but had not time to finish it. So maybe if you added more models, or other things, this might have been the issue as well.", "1035534": "Congratulations \nAwesome work Thanks for sharing your solution overview with us", "1037249": "Congratulations to all the hard working people.", "1493221": "congratulations!" }, "source": "meta" }