{ "id": 316959, "title": "Interresting Paper: A Metric Learning Reality Check", "url": "/competitions/happy-whale-and-dolphin/discussion/316959", "author_name": "", "post_date": "2022-04-04T20:30:22.678323400Z", "votes": 12, "comment_count": 6, "views": 0, "content": "

When searching for interresting papers that could provide some further information that could be useful for this competition a came across a very interresting paper:

\n

A Metric Learning Reality Check

\n

As the summary states a lot of recent metric learning papers seem to set new state of the art results with almost every paper that is released.

\n

This paper dives into comparing all those metric learning papers. In a very thorough review some common mistakes/flaws in those papers are analysed and finally a very extensive and complete review is performed on all those metric learning models.

\n

With a common setup all metric learning models are compared across multiple datasets in the same comparable way. The reality turns out to be that progress over the last 15 years is not that great as some papers want us to believe.

\n

Fun fact: ArcFace turns out to have a rather strong performance.

\n

I hope this paper provides you all with some interresting insights.

", "messages": [ { "id": "1745310", "postDate": "04/04/2022 20:30:22", "content": "

When searching for interresting papers that could provide some further information that could be useful for this competition a came across a very interresting paper:

\n

A Metric Learning Reality Check

\n

As the summary states a lot of recent metric learning papers seem to set new state of the art results with almost every paper that is released.

\n

This paper dives into comparing all those metric learning papers. In a very thorough review some common mistakes/flaws in those papers are analysed and finally a very extensive and complete review is performed on all those metric learning models.

\n

With a common setup all metric learning models are compared across multiple datasets in the same comparable way. The reality turns out to be that progress over the last 15 years is not that great as some papers want us to believe.

\n

Fun fact: ArcFace turns out to have a rather strong performance.

\n

I hope this paper provides you all with some interresting insights.

", "rawMarkdown": "When searching for interresting papers that could provide some further information that could be useful for this competition a came across a very interresting paper:\n\n[A Metric Learning Reality Check](https://arxiv.org/abs/2003.08505v3)\n\nAs the summary states a lot of recent metric learning papers seem to set new state of the art results with almost every paper that is released.\n\nThis paper dives into comparing all those metric learning papers. In a very thorough review some common mistakes/flaws in those papers are analysed and finally a very extensive and complete review is performed on all those metric learning models.\n\nWith a common setup all metric learning models are compared across multiple datasets in the same comparable way. The reality turns out to be that progress over the last 15 years is not that great as some papers want us to believe.\n\nFun fact: ArcFace turns out to have a rather strong performance.\n\nI hope this paper provides you all with some interresting insights.", "votes": null }, { "id": "1745383", "postDate": "04/04/2022 22:22:45", "content": "

Thank you ！Nice to hava papers to read.

", "rawMarkdown": "Thank you ！Nice to hava papers to read.", "votes": null }, { "id": "1745956", "postDate": "04/05/2022 12:42:13", "content": "

Thanks @rsmits defnintely a gd read summarising all those papers

", "rawMarkdown": "Thanks @rsmits defnintely a gd read summarising all those papers", "votes": null }, { "id": "1746284", "postDate": "04/05/2022 16:21:26", "content": "

Thank you for the article, SoftTriple looks worthy to study

", "rawMarkdown": "Thank you for the article, SoftTriple looks worthy to study", "votes": null }, { "id": "1746498", "postDate": "04/05/2022 21:20:09", "content": "

Hi @alanhabrony There are plenty of papers…but this one is really worth the reading time :-)

", "rawMarkdown": "Hi @alanhabrony There are plenty of papers...but this one is really worth the reading time :-)", "votes": null }, { "id": "1746510", "postDate": "04/05/2022 21:54:29", "content": "

You're welcome @samsonlo

", "rawMarkdown": "You're welcome @samsonlo", "votes": null }, { "id": "1746513", "postDate": "04/05/2022 22:02:52", "content": "

Hi @kwentar Yes that one also looks promising. Looking forward to your write up after the competition deadline. Good luck!

", "rawMarkdown": "Hi @kwentar Yes that one also looks promising. Looking forward to your write up after the competition deadline. Good luck!", "votes": null } ], "comments": [ { "id": 1745383, "author_name": "alanhabrony", "author_url": "", "post_date": "04/04/2022 22:22:45", "content": "

Thank you ！Nice to hava papers to read.

", "votes": null, "replies": [ { "id": 1746498, "author_name": "rsmits", "author_url": "", "post_date": "04/05/2022 21:20:09", "content": "

Hi @alanhabrony There are plenty of papers…but this one is really worth the reading time :-)

", "votes": null, "replies": [] } ] }, { "id": 1745956, "author_name": "samsonlo", "author_url": "", "post_date": "04/05/2022 12:42:13", "content": "

Thanks @rsmits defnintely a gd read summarising all those papers

", "votes": null, "replies": [ { "id": 1746510, "author_name": "rsmits", "author_url": "", "post_date": "04/05/2022 21:54:29", "content": "

You're welcome @samsonlo

", "votes": null, "replies": [] } ] }, { "id": 1746284, "author_name": "kwentar", "author_url": "", "post_date": "04/05/2022 16:21:26", "content": "

Thank you for the article, SoftTriple looks worthy to study

", "votes": null, "replies": [ { "id": 1746513, "author_name": "rsmits", "author_url": "", "post_date": "04/05/2022 22:02:52", "content": "

Hi @kwentar Yes that one also looks promising. Looking forward to your write up after the competition deadline. Good luck!

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1745310": "When searching for interresting papers that could provide some further information that could be useful for this competition a came across a very interresting paper:\n\n[A Metric Learning Reality Check](https://arxiv.org/abs/2003.08505v3)\n\nAs the summary states a lot of recent metric learning papers seem to set new state of the art results with almost every paper that is released.\n\nThis paper dives into comparing all those metric learning papers. In a very thorough review some common mistakes/flaws in those papers are analysed and finally a very extensive and complete review is performed on all those metric learning models.\n\nWith a common setup all metric learning models are compared across multiple datasets in the same comparable way. The reality turns out to be that progress over the last 15 years is not that great as some papers want us to believe.\n\nFun fact: ArcFace turns out to have a rather strong performance.\n\nI hope this paper provides you all with some interresting insights.", "1745383": "Thank you ！Nice to hava papers to read.", "1745956": "Thanks @rsmits defnintely a gd read summarising all those papers", "1746284": "Thank you for the article, SoftTriple looks worthy to study", "1746498": "Hi @alanhabrony There are plenty of papers...but this one is really worth the reading time :-)", "1746510": "You're welcome @samsonlo", "1746513": "Hi @kwentar Yes that one also looks promising. Looking forward to your write up after the competition deadline. Good luck!" }, "source": "meta" }