{ "id": 410406, "title": "Is Kaggle CPU and disk too slow?", "url": "/competitions/image-matching-challenge-2023/discussion/410406", "author_name": "", "post_date": "2023-05-15T07:53:53.193999200Z", "votes": null, "comment_count": 2, "views": 0, "content": "

In my local machine with AMD Ryzen 9 5950X 16-Core Processor and SSD, it takes about 20-30 seconds to reconstruct the bike scene while it takes more than 3 minutes on Kaggle. We also can't use all images on larger scenes because execution time grows quadratically. Maybe this competition shouldn't be a code competition after all, what do you think?

", "messages": [ { "id": "2259761", "postDate": "05/15/2023 07:53:53", "content": "

In my local machine with AMD Ryzen 9 5950X 16-Core Processor and SSD, it takes about 20-30 seconds to reconstruct the bike scene while it takes more than 3 minutes on Kaggle. We also can't use all images on larger scenes because execution time grows quadratically. Maybe this competition shouldn't be a code competition after all, what do you think?

", "rawMarkdown": "In my local machine with AMD Ryzen 9 5950X 16-Core Processor and SSD, it takes about 20-30 seconds to reconstruct the bike scene while it takes more than 3 minutes on Kaggle. We also can't use all images on larger scenes because execution time grows quadratically. Maybe this competition shouldn't be a code competition after all, what do you think?", "votes": null }, { "id": "2259782", "postDate": "05/15/2023 08:09:44", "content": "

The aim of the competition is to propose new and original solutions for challenging problems: https://www.kaggle.com/competitions/image-matching-challenge-2023/overview

\n

This includes new implementations and approaches to the solution. The aim of the challenge is not \"ok here there are colmap, LoFTR and SuperGlue, just combine and tune them\".

\n

Anyways, please try to look at previous discussions: https://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/406927

", "rawMarkdown": "The aim of the competition is to propose new and original solutions for challenging problems: https://www.kaggle.com/competitions/image-matching-challenge-2023/overview\n\nThis includes new implementations and approaches to the solution. The aim of the challenge is not \"ok here there are colmap, LoFTR and SuperGlue, just combine and tune them\".\n\nAnyways, please try to look at previous discussions: https://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/406927", "votes": null }, { "id": "2259795", "postDate": "05/15/2023 08:20:41", "content": "

The point is to put everyone on relatively even footing. You can always throw endless amounts of compute at it, but that won't teach us much.

\n

We know that the problem can be solved within these constraints: there are pretty strong solutions in the leaderboards already.

", "rawMarkdown": "The point is to put everyone on relatively even footing. You can always throw endless amounts of compute at it, but that won't teach us much.\n\nWe know that the problem can be solved within these constraints: there are pretty strong solutions in the leaderboards already.", "votes": null } ], "comments": [ { "id": 2259782, "author_name": "fabiobellavia", "author_url": "", "post_date": "05/15/2023 08:09:44", "content": "

The aim of the competition is to propose new and original solutions for challenging problems: https://www.kaggle.com/competitions/image-matching-challenge-2023/overview

\n

This includes new implementations and approaches to the solution. The aim of the challenge is not \"ok here there are colmap, LoFTR and SuperGlue, just combine and tune them\".

\n

Anyways, please try to look at previous discussions: https://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/406927

", "votes": null, "replies": [] }, { "id": 2259795, "author_name": "eduardtrulls", "author_url": "", "post_date": "05/15/2023 08:20:41", "content": "

The point is to put everyone on relatively even footing. You can always throw endless amounts of compute at it, but that won't teach us much.

\n

We know that the problem can be solved within these constraints: there are pretty strong solutions in the leaderboards already.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2259761": "In my local machine with AMD Ryzen 9 5950X 16-Core Processor and SSD, it takes about 20-30 seconds to reconstruct the bike scene while it takes more than 3 minutes on Kaggle. We also can't use all images on larger scenes because execution time grows quadratically. Maybe this competition shouldn't be a code competition after all, what do you think?", "2259782": "The aim of the competition is to propose new and original solutions for challenging problems: https://www.kaggle.com/competitions/image-matching-challenge-2023/overview\n\nThis includes new implementations and approaches to the solution. The aim of the challenge is not \"ok here there are colmap, LoFTR and SuperGlue, just combine and tune them\".\n\nAnyways, please try to look at previous discussions: https://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/406927", "2259795": "The point is to put everyone on relatively even footing. You can always throw endless amounts of compute at it, but that won't teach us much.\n\nWe know that the problem can be solved within these constraints: there are pretty strong solutions in the leaderboards already." }, "source": "meta" }