{ "id": 266367, "title": "Compare Landmark Recognition 2021 vs 2020 ?", "url": "/competitions/landmark-recognition-2021/discussion/266367", "author_name": "SeshuRaju 🧘‍♂️", "post_date": "2021-08-18T21:08:57.574000", "votes": 1, "comment_count": 3, "views": 0, "content": "

For this year, we introduce a new test set which is sampled from many countries, increasing the diversity in worldwide representation.

\n

Since training data not changed as per Landmark retrieval / recognition data compare!!

\n

is new hidden test data is a completely different distribution than train data. is the reason for last year's trained models won't work on this year's test data?

\n

Please correct me if I wrong in understanding this year goal

\n

Note: I understand this year challenge compare to last year

\n", "messages": [ { "id": 1480196, "postDate": "2021-08-18T21:08:57.573Z", "content": "

For this year, we introduce a new test set which is sampled from many countries, increasing the diversity in worldwide representation.

\n

Since training data not changed as per Landmark retrieval / recognition data compare!!

\n

is new hidden test data is a completely different distribution than train data. is the reason for last year's trained models won't work on this year's test data?

\n

Please correct me if I wrong in understanding this year goal

\n

Note: I understand this year challenge compare to last year

\n", "rawMarkdown": "**For this year, we introduce a new test set which is sampled from many countries, increasing the diversity in worldwide representation.**\n\nSince training data not changed as per [Landmark retrieval / recognition data compare](https://www.kaggle.com/inoueu1/landmark-retrieval-recognition-data-compare)!!\n\nis new hidden test data is a completely different distribution than train data. is the reason for last year's trained models won't work on this year's test data?\n\nPlease correct me if I wrong in understanding this year goal\n\n**Note**: I understand this year challenge compare to last year \n- https://www.kaggle.com/c/landmark-recognition-2021/discussion/266416 ( this will give an idea what changed !!! )", "votes": 1 }, { "id": 1480690, "postDate": "2021-08-19T06:22:47.650Z", "content": "
\n

is the reason for last year's trained models won't work on this year's test data?

\n
\n

do you already know that?

", "rawMarkdown": "> is the reason for last year's trained models won't work on this year's test data?\n\ndo you already know that?", "replies": [ { "id": 1481004, "postDate": "2021-08-19T09:08:50.150Z", "content": "

Just a guess, since last year top solution team's scores are started from 0.2+ as an observation ( I expected last year trained models work well since training data not changed ).
\n2020: Dieter > bestfitting > Ahmet
\n2021: Ahmet > bestfitting > Dieter

\n

But not able to understand this year goal compare to last year. the new test dataset ( is it means, it is too hard to find landmarks in new test dataset ? ) @philippsinger

", "rawMarkdown": "Just a guess, since last year top solution team's scores are started from 0.2+ as an observation ( I expected last year trained models work well since training data not changed ). \n2020: **Dieter** > bestfitting > Ahmet\n2021: **Ahmet** > bestfitting > Dieter\n\nBut not able to understand this year goal compare to last year. the new test dataset ( is it means, it is too hard to find landmarks in new test dataset ? ) @philippsinger\n\n" } ] }, { "id": 1481249, "postDate": "2021-08-19T11:43:41.310Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1480690, "author_name": "Psi", "author_url": "", "post_date": "2021-08-19T06:22:47.650000", "content": "
\n

is the reason for last year's trained models won't work on this year's test data?

\n
\n

do you already know that?

", "votes": 0, "replies": [ { "id": 1481004, "author_name": "SeshuRaju 🧘‍♂️", "author_url": "", "post_date": "2021-08-19T09:08:50.150000", "content": "

Just a guess, since last year top solution team's scores are started from 0.2+ as an observation ( I expected last year trained models work well since training data not changed ).
\n2020: Dieter > bestfitting > Ahmet
\n2021: Ahmet > bestfitting > Dieter

\n

But not able to understand this year goal compare to last year. the new test dataset ( is it means, it is too hard to find landmarks in new test dataset ? ) @philippsinger

", "votes": 0, "replies": [] } ] }, { "id": 1481249, "author_name": "", "author_url": "", "post_date": "2021-08-19T11:43:41.310000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1480196": "**For this year, we introduce a new test set which is sampled from many countries, increasing the diversity in worldwide representation.**\n\nSince training data not changed as per [Landmark retrieval / recognition data compare](https://www.kaggle.com/inoueu1/landmark-retrieval-recognition-data-compare)!!\n\nis new hidden test data is a completely different distribution than train data. is the reason for last year's trained models won't work on this year's test data?\n\nPlease correct me if I wrong in understanding this year goal\n\n**Note**: I understand this year challenge compare to last year \n- https://www.kaggle.com/c/landmark-recognition-2021/discussion/266416 ( this will give an idea what changed !!! )", "1480690": "> is the reason for last year's trained models won't work on this year's test data?\n\ndo you already know that?", "1481249": "" } }