{ "id": 304504, "title": "Previous Happywhale Competition Solutions", "url": "/competitions/happy-whale-and-dolphin/discussion/304504", "author_name": "", "post_date": "2022-02-01T16:46:43.323495400Z", "votes": 105, "comment_count": 17, "views": 0, "content": "", "messages": [ { "id": "1671681", "postDate": "02/01/2022 16:46:43", "content": "", "rawMarkdown": "[1st solution(classification) && code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82366) by @qiaojian.\n- [2nd place code, end to end whale Identification model](https://www.kaggle.com/c/humpback-whale-identification/discussion/83885) by @shentao.\n- [3rd place solution with code: ArcFace](https://www.kaggle.com/c/humpback-whale-identification/discussion/82484) by @pudae81.\n- [4th Place Solution: SIFT + Siamese](https://www.kaggle.com/c/humpback-whale-identification/discussion/82356) by @tivfrvqhs5.\n- [5th solution blog post + code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82369) by @weimin.\n- [7th place Pure Magic thanks Radek solution: classification](https://www.kaggle.com/c/humpback-whale-identification/discussion/82352) by @oldufo.\n- [9th place solution or how we spent last one and a half month](https://www.kaggle.com/c/humpback-whale-identification/discussion/82427) by @sawseen.\n- [10th Place Solution: Metric Learning, Classification, Siamese, LGBM](https://www.kaggle.com/c/humpback-whale-identification/discussion/82430) by @vshakhray.\n- [11th golden place. Special thanks](https://www.kaggle.com/c/humpback-whale-identification/discussion/82415) by @abmokin.\n- [15-th plcae solution: sphereface, image alignment and multi-layer fusion](https://www.kaggle.com/c/humpback-whale-identification/discussion/82361) by @xf1994.\n- [24th place solution](https://www.kaggle.com/c/humpback-whale-identification/discussion/82359) by @interneuron.\n- [25-th place solution: CosFace + ProtoNets](https://www.kaggle.com/c/humpback-whale-identification/discussion/82409) by @melgor.\n- [27th Place solution (pure metric, pure C++)](https://www.kaggle.com/c/humpback-whale-identification/discussion/82423) by @pinullmezon.\n- [Approximation solution for LAP, 30th place solution](https://www.kaggle.com/c/humpback-whale-identification/discussion/82380) by @mingzhao03.\n- [31st place solution + source code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82393) by @suicaokhoailang.\n- [57th place solution, SoftTripletLoss, 256x512 image size, fastai v1](https://www.kaggle.com/c/humpback-whale-identification/discussion/82364) by @mnpinto.\n- [87 place solution - thanks Martin](https://www.kaggle.com/c/humpback-whale-identification/discussion/82994) by @spsancti.\n- [[143th place] single model trained with cross entropy / contrastive loss using fastai](https://www.kaggle.com/c/humpback-whale-identification/discussion/82480) by @radek1.", "votes": null }, { "id": "1671755", "postDate": "02/01/2022 17:47:07", "content": "

Nice! We implemented the 3rd place solution from @pudae81 into Happywhale and published about the result.

", "rawMarkdown": "Nice! We implemented the 3rd place solution from @pudae81 into [Happywhale](https://happywhale.com/) and [published about the result](https://rdcu.be/cCOtw).", "votes": null }, { "id": "1671788", "postDate": "02/01/2022 18:17:17", "content": "

Thanks for your post!

", "rawMarkdown": "Thanks for your post!", "votes": null }, { "id": "1671911", "postDate": "02/01/2022 19:25:48", "content": "

Here is the notebook that explains the metric, by @pestipeti
\nhttps://www.kaggle.com/pestipeti/explanation-of-map5-scoring-metric/notebook

", "rawMarkdown": "Here is the notebook that explains the metric, by @pestipeti \nhttps://www.kaggle.com/pestipeti/explanation-of-map5-scoring-metric/notebook", "votes": null }, { "id": "1675193", "postDate": "02/04/2022 05:05:58", "content": "

Irrelevant question but why did you selected the 3rd place solution for implementation and not the 1st or 2nd or someother

", "rawMarkdown": "Irrelevant question but why did you selected the 3rd place solution for implementation and not the 1st or 2nd or someother", "votes": null }, { "id": "1675219", "postDate": "02/04/2022 05:39:31", "content": "

Not irrelevant at all, in fact something I am intending to post about asap. @pudae81's solution was selected for implementation because it was good clean well commented code that did a solid job of separating processes, and used computer processor resources efficiently. The tendency here is towards competition code, which can be a bit messy, but we are looking for something highly implementable in applied marine conservation science, with a very small programming team behind Happywhale.

\n

The state of the code is not something that winning is dependent on but hopefully competitors are keen to see their results implemented and making a real difference in the world, so for that, clean, well commented and well segmented code that efficiently uses computer processor resources, these are the golden eggs we hope come out of the competition :-)

", "rawMarkdown": "Not irrelevant at all, in fact something I am intending to post about asap. @pudae81's solution was selected for implementation because it was good clean well commented code that did a solid job of separating processes, and used computer processor resources efficiently. The tendency here is towards competition code, which can be a bit messy, but we are looking for something highly implementable in applied marine conservation science, with a very small programming team behind [Happywhale](https://happywhale.com/).\n\nThe state of the code is not something that winning is dependent on but hopefully competitors are keen to see their results implemented and making a real difference in the world, so for that, clean, well commented and well segmented code that efficiently uses computer processor resources, these are the golden eggs we hope come out of the competition :-)", "votes": null }, { "id": "1676647", "postDate": "02/05/2022 06:54:18", "content": "

very reasonable, I think this is why this isn't a code competition.
\nmany talented people here, I am sure you will get plenty of well-documented code with high scores in this comp.

", "rawMarkdown": "very reasonable, I think this is why this isn't a code competition.\nmany talented people here, I am sure you will get plenty of well-documented code with high scores in this comp.", "votes": null }, { "id": "1676684", "postDate": "02/05/2022 07:50:22", "content": "

Another question, have you tried training this new data with the winning models of the last comp? how were they performing? Are they performing well on new data?
\nSince the code is public, I fear someone using the exact previous codes and get high score.

", "rawMarkdown": "Another question, have you tried training this new data with the winning models of the last comp? how were they performing? Are they performing well on new data?\nSince the code is public, I fear someone using the exact previous codes and get high score.", "votes": null }, { "id": "1676995", "postDate": "02/05/2022 12:45:56", "content": "

The code from the previous top 5 are public by competition rules and others have posted their code as well. Nothing wrong with using pieces from that but it definitely won't work well without substantial further development.

", "rawMarkdown": "The code from the previous top 5 are public by competition rules and others have posted their code as well. Nothing wrong with using pieces from that but it definitely won't work well without substantial further development.", "votes": null }, { "id": "1677636", "postDate": "02/05/2022 21:25:03", "content": "

Thanks for sharing this :)

", "rawMarkdown": "Thanks for sharing this :)", "votes": null }, { "id": "1678903", "postDate": "02/06/2022 20:53:44", "content": "

Thank you for your work to get those solutions together.

", "rawMarkdown": "Thank you for your work to get those solutions together.", "votes": null }, { "id": "1680148", "postDate": "02/07/2022 15:56:02", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": null }, { "id": "1683122", "postDate": "02/09/2022 15:33:38", "content": "

Thanks for sharing, very useful!

", "rawMarkdown": "Thanks for sharing, very useful!", "votes": null }, { "id": "1700178", "postDate": "02/21/2022 18:11:08", "content": "

Thank you very much for sharing!

", "rawMarkdown": "Thank you very much for sharing!", "votes": null }, { "id": "1703058", "postDate": "02/24/2022 06:53:59", "content": "

Great post! It would be even greater if each link had (tf/keras, pytorch, etc, annotation). For example, the 3rd place was done in pytorch.

", "rawMarkdown": "Great post! It would be even greater if each link had (tf/keras, pytorch, etc, annotation). For example, the 3rd place was done in pytorch.", "votes": null }, { "id": "1755696", "postDate": "04/14/2022 21:39:52", "content": "

Interesting

", "rawMarkdown": "Interesting", "votes": null }, { "id": "1755824", "postDate": "04/15/2022 03:09:07", "content": "

Thank you for solution bookmark. This very helpful

", "rawMarkdown": "Thank you for solution bookmark. This very helpful", "votes": null }, { "id": "1756444", "postDate": "04/15/2022 14:21:53", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": null } ], "comments": [ { "id": 1671755, "author_name": "tedcheese", "author_url": "", "post_date": "02/01/2022 17:47:07", "content": "

Nice! We implemented the 3rd place solution from @pudae81 into Happywhale and published about the result.

", "votes": null, "replies": [ { "id": 1675193, "author_name": "mrinath", "author_url": "", "post_date": "02/04/2022 05:05:58", "content": "

Irrelevant question but why did you selected the 3rd place solution for implementation and not the 1st or 2nd or someother

", "votes": null, "replies": [] }, { "id": 1675219, "author_name": "tedcheese", "author_url": "", "post_date": "02/04/2022 05:39:31", "content": "

Not irrelevant at all, in fact something I am intending to post about asap. @pudae81's solution was selected for implementation because it was good clean well commented code that did a solid job of separating processes, and used computer processor resources efficiently. The tendency here is towards competition code, which can be a bit messy, but we are looking for something highly implementable in applied marine conservation science, with a very small programming team behind Happywhale.

\n

The state of the code is not something that winning is dependent on but hopefully competitors are keen to see their results implemented and making a real difference in the world, so for that, clean, well commented and well segmented code that efficiently uses computer processor resources, these are the golden eggs we hope come out of the competition :-)

", "votes": null, "replies": [] }, { "id": 1676647, "author_name": "mrinath", "author_url": "", "post_date": "02/05/2022 06:54:18", "content": "

very reasonable, I think this is why this isn't a code competition.
\nmany talented people here, I am sure you will get plenty of well-documented code with high scores in this comp.

", "votes": null, "replies": [] }, { "id": 1676684, "author_name": "mrinath", "author_url": "", "post_date": "02/05/2022 07:50:22", "content": "

Another question, have you tried training this new data with the winning models of the last comp? how were they performing? Are they performing well on new data?
\nSince the code is public, I fear someone using the exact previous codes and get high score.

", "votes": null, "replies": [] }, { "id": 1676995, "author_name": "tedcheese", "author_url": "", "post_date": "02/05/2022 12:45:56", "content": "

The code from the previous top 5 are public by competition rules and others have posted their code as well. Nothing wrong with using pieces from that but it definitely won't work well without substantial further development.

", "votes": null, "replies": [] } ] }, { "id": 1671788, "author_name": "w13vitaliy", "author_url": "", "post_date": "02/01/2022 18:17:17", "content": "

Thanks for your post!

", "votes": null, "replies": [] }, { "id": 1671911, "author_name": "ipythonx", "author_url": "", "post_date": "02/01/2022 19:25:48", "content": "

Here is the notebook that explains the metric, by @pestipeti
\nhttps://www.kaggle.com/pestipeti/explanation-of-map5-scoring-metric/notebook

", "votes": null, "replies": [] }, { "id": 1677636, "author_name": "ammarabbasi1040", "author_url": "", "post_date": "02/05/2022 21:25:03", "content": "

Thanks for sharing this :)

", "votes": null, "replies": [] }, { "id": 1678903, "author_name": "nebipeker", "author_url": "", "post_date": "02/06/2022 20:53:44", "content": "

Thank you for your work to get those solutions together.

", "votes": null, "replies": [] }, { "id": 1680148, "author_name": "sergeyzin", "author_url": "", "post_date": "02/07/2022 15:56:02", "content": "

Thanks for sharing!

", "votes": null, "replies": [] }, { "id": 1683122, "author_name": "saurabhbagchi", "author_url": "", "post_date": "02/09/2022 15:33:38", "content": "

Thanks for sharing, very useful!

", "votes": null, "replies": [] }, { "id": 1700178, "author_name": "datascientistfp", "author_url": "", "post_date": "02/21/2022 18:11:08", "content": "

Thank you very much for sharing!

", "votes": null, "replies": [] }, { "id": 1703058, "author_name": "dmitrykonovalov", "author_url": "", "post_date": "02/24/2022 06:53:59", "content": "

Great post! It would be even greater if each link had (tf/keras, pytorch, etc, annotation). For example, the 3rd place was done in pytorch.

", "votes": null, "replies": [] }, { "id": 1755696, "author_name": "luketambakis", "author_url": "", "post_date": "04/14/2022 21:39:52", "content": "

Interesting

", "votes": null, "replies": [] }, { "id": 1755824, "author_name": "chocolatemilk", "author_url": "", "post_date": "04/15/2022 03:09:07", "content": "

Thank you for solution bookmark. This very helpful

", "votes": null, "replies": [] }, { "id": 1756444, "author_name": "sayanmu", "author_url": "", "post_date": "04/15/2022 14:21:53", "content": "

Thanks for sharing!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1671681": "[1st solution(classification) && code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82366) by @qiaojian.\n- [2nd place code, end to end whale Identification model](https://www.kaggle.com/c/humpback-whale-identification/discussion/83885) by @shentao.\n- [3rd place solution with code: ArcFace](https://www.kaggle.com/c/humpback-whale-identification/discussion/82484) by @pudae81.\n- [4th Place Solution: SIFT + Siamese](https://www.kaggle.com/c/humpback-whale-identification/discussion/82356) by @tivfrvqhs5.\n- [5th solution blog post + code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82369) by @weimin.\n- [7th place Pure Magic thanks Radek solution: classification](https://www.kaggle.com/c/humpback-whale-identification/discussion/82352) by @oldufo.\n- [9th place solution or how we spent last one and a half month](https://www.kaggle.com/c/humpback-whale-identification/discussion/82427) by @sawseen.\n- [10th Place Solution: Metric Learning, Classification, Siamese, LGBM](https://www.kaggle.com/c/humpback-whale-identification/discussion/82430) by @vshakhray.\n- [11th golden place. Special thanks](https://www.kaggle.com/c/humpback-whale-identification/discussion/82415) by @abmokin.\n- [15-th plcae solution: sphereface, image alignment and multi-layer fusion](https://www.kaggle.com/c/humpback-whale-identification/discussion/82361) by @xf1994.\n- [24th place solution](https://www.kaggle.com/c/humpback-whale-identification/discussion/82359) by @interneuron.\n- [25-th place solution: CosFace + ProtoNets](https://www.kaggle.com/c/humpback-whale-identification/discussion/82409) by @melgor.\n- [27th Place solution (pure metric, pure C++)](https://www.kaggle.com/c/humpback-whale-identification/discussion/82423) by @pinullmezon.\n- [Approximation solution for LAP, 30th place solution](https://www.kaggle.com/c/humpback-whale-identification/discussion/82380) by @mingzhao03.\n- [31st place solution + source code](https://www.kaggle.com/c/humpback-whale-identification/discussion/82393) by @suicaokhoailang.\n- [57th place solution, SoftTripletLoss, 256x512 image size, fastai v1](https://www.kaggle.com/c/humpback-whale-identification/discussion/82364) by @mnpinto.\n- [87 place solution - thanks Martin](https://www.kaggle.com/c/humpback-whale-identification/discussion/82994) by @spsancti.\n- [[143th place] single model trained with cross entropy / contrastive loss using fastai](https://www.kaggle.com/c/humpback-whale-identification/discussion/82480) by @radek1.", "1671755": "Nice! We implemented the 3rd place solution from @pudae81 into [Happywhale](https://happywhale.com/) and [published about the result](https://rdcu.be/cCOtw).", "1671788": "Thanks for your post!", "1671911": "Here is the notebook that explains the metric, by @pestipeti \nhttps://www.kaggle.com/pestipeti/explanation-of-map5-scoring-metric/notebook", "1675193": "Irrelevant question but why did you selected the 3rd place solution for implementation and not the 1st or 2nd or someother", "1675219": "Not irrelevant at all, in fact something I am intending to post about asap. @pudae81's solution was selected for implementation because it was good clean well commented code that did a solid job of separating processes, and used computer processor resources efficiently. The tendency here is towards competition code, which can be a bit messy, but we are looking for something highly implementable in applied marine conservation science, with a very small programming team behind [Happywhale](https://happywhale.com/).\n\nThe state of the code is not something that winning is dependent on but hopefully competitors are keen to see their results implemented and making a real difference in the world, so for that, clean, well commented and well segmented code that efficiently uses computer processor resources, these are the golden eggs we hope come out of the competition :-)", "1676647": "very reasonable, I think this is why this isn't a code competition.\nmany talented people here, I am sure you will get plenty of well-documented code with high scores in this comp.", "1676684": "Another question, have you tried training this new data with the winning models of the last comp? how were they performing? Are they performing well on new data?\nSince the code is public, I fear someone using the exact previous codes and get high score.", "1676995": "The code from the previous top 5 are public by competition rules and others have posted their code as well. Nothing wrong with using pieces from that but it definitely won't work well without substantial further development.", "1677636": "Thanks for sharing this :)", "1678903": "Thank you for your work to get those solutions together.", "1680148": "Thanks for sharing!", "1683122": "Thanks for sharing, very useful!", "1700178": "Thank you very much for sharing!", "1703058": "Great post! It would be even greater if each link had (tf/keras, pytorch, etc, annotation). For example, the 3rd place was done in pytorch.", "1755696": "Interesting", "1755824": "Thank you for solution bookmark. This very helpful", "1756444": "Thanks for sharing!" }, "source": "meta" }