We noticed in last year's challenge that some participants relied on code with non-commercial licenses for their solutions: specifically, the official implementations of SuperPoint and SuperGlue. Unfortunately this made a few top solutions unable to claim prizes.
\nWhile you are ultimately responsible to ensure that your submission complies with the competition rules, we will provide guidance if possible. For example, there are other implementations of SuperPoint. We also hope to be able to point you to a new implementation of SuperGlue soon.)
", "messages": [ { "id": 2218588, "postDate": "2023-04-11T20:24:58.520Z", "content": "We noticed in last year's challenge that some participants relied on code with non-commercial licenses for their solutions: specifically, the official implementations of SuperPoint and SuperGlue. Unfortunately this made a few top solutions unable to claim prizes.
\nWhile you are ultimately responsible to ensure that your submission complies with the competition rules, we will provide guidance if possible. For example, there are other implementations of SuperPoint. We also hope to be able to point you to a new implementation of SuperGlue soon.)
", "rawMarkdown": "We noticed in [last year's challenge](https://www.kaggle.com/competitions/image-matching-challenge-2022/overview) that some participants relied on code with non-commercial licenses for their solutions: specifically, the official implementations of [SuperPoint](https://github.com/magicleap/SuperPointPretrainedNetwork) and [SuperGlue](https://github.com/magicleap/SuperGluePretrainedNetwork). Unfortunately this made a few top solutions unable to claim prizes.\n\nWhile you are ultimately responsible to ensure that your submission complies with the [competition rules](https://www.kaggle.com/competitions/image-matching-challenge-2023/rules), we will provide guidance if possible. For example, there are other implementations of [SuperPoint](https://github.com/rpautrat/SuperPoint). We also hope to be able to point you to a new implementation of SuperGlue soon.)", "votes": 17 }, { "id": 2218649, "postDate": "2023-04-11T21:59:16.427Z", "content": "\n\nUnfortunately this made a few top solutions unable to claim prizes.
\n
Thats quite an understatement. We were 12th place, yet got prize of 2nd place. Which means 10 teams before us were not eligible for the prize. Or if you see it differently, we followed the rules and got the 2nd best prize eligible solution, yet not even got a gold medal…
", "rawMarkdown": ">Unfortunately this made a few top solutions unable to claim prizes.\n\nThats quite an understatement. We were 12th place, yet got prize of 2nd place. Which means 10 teams before us were not eligible for the prize. Or if you see it differently, we followed the rules and got the 2nd best prize eligible solution, yet not even got a gold medal...", "votes": 10, "replies": [ { "id": 2218654, "postDate": "2023-04-11T22:09:46.303Z", "content": "Well, the rules allowed non-open submission \"for glory only\", so 10 teams above you also followed rules. Which, arguable, were not ideal, but that's life. However we don't expect SG submissions to dominate this year.
", "rawMarkdown": "Well, the rules allowed non-open submission \"for glory only\", so 10 teams above you also followed rules. Which, arguable, were not ideal, but that's life. However we don't expect SG submissions to dominate this year. " }, { "id": 2223640, "postDate": "2023-04-16T13:33:48.443Z", "content": "True :)
\n2nd but without gold 😂 I remember.
If non-commercial code leads to higher ranks but doesn't allow to claim money prize, does it mean that each team have to decide what they are competing for? I'm speaking mostly about situation from last year's challenge, but the scenario could be the same this year. Could you please clarify one moment, will you judge only one submission? What if I submit one with higher score but with non-commercial code and second one with lower score but allowed license. Can I get gold based on first submit and prize based on second together? Thanks!
", "rawMarkdown": "If non-commercial code leads to higher ranks but doesn't allow to claim money prize, does it mean that each team have to decide what they are competing for? I'm speaking mostly about situation from last year's challenge, but the scenario could be the same this year. Could you please clarify one moment, will you judge only one submission? What if I submit one with higher score but with non-commercial code and second one with lower score but allowed license. Can I get gold based on first submit and prize based on second together? Thanks!", "votes": 1, "replies": [ { "id": 2234034, "postDate": "2023-04-24T19:44:14.460Z", "content": "We're only going to review the highest scoring submission from each team.
", "rawMarkdown": "We're only going to review the highest scoring submission from each team.", "votes": 1 } ] }, { "id": 2219704, "postDate": "2023-04-12T19:53:58.820Z", "content": "@eduardtrulls @oldufo hello, I didn't fully understand requirements to be invited to the CVPR workshop. Should we win a gold model, should we be a price contender?
", "rawMarkdown": "@eduardtrulls @oldufo hello, I didn't fully understand requirements to be invited to the CVPR workshop. Should we win a gold model, should we be a price contender? ", "votes": 1, "replies": [ { "id": 2219715, "postDate": "2023-04-12T20:04:16.030Z", "content": "The result should be good enough, and the solution should be interesting enough. Usually being in top-10 is good enough, but this doesn't mean that we automatically invite all top-10 teams, or that 20th place cannot be invited.
\nUnlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition.
", "rawMarkdown": "The result should be good enough, and the solution should be interesting enough. Usually being in top-10 is good enough, but this doesn't mean that we automatically invite all top-10 teams, or that 20th place cannot be invited. \n\nUnlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition.", "votes": 3, "replies": [ { "id": 2219775, "postDate": "2023-04-12T21:19:23.633Z", "content": "Should the solution be among the chosen (in kaggle for private scoring) to be considered for the workshop?
", "rawMarkdown": "Should the solution be among the chosen (in kaggle for private scoring) to be considered for the workshop? " }, { "id": 2219779, "postDate": "2023-04-12T21:23:54.670Z", "content": "I can only repeat why I have already said:
\n\n", "rawMarkdown": "I can only repeat why I have already said:\n\n>Unlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition." } ] } ] }, { "id": 2321576, "postDate": "2023-06-28T17:16:11.733Z", "content": "Unlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition.
\n
Is there any update on the prize winners?
", "rawMarkdown": "Is there any update on the prize winners?" }, { "id": 2300925, "postDate": "2023-06-13T14:16:01.790Z", "content": "What are the five teams that got the bonus this year?
", "rawMarkdown": "What are the five teams that got the bonus this year?", "replies": [ { "id": 2301111, "postDate": "2023-06-13T16:15:03.760Z", "content": "After the result verification by Kaggle team, they and we will reach out starting from the top, and checking if the solution is prize-eligible. As a rule of thumb - any MagicLeap code/weights -- such as SuperPoint and/or SuperGlue disqualifies a team from money due to license restriction.
", "rawMarkdown": "After the result verification by Kaggle team, they and we will reach out starting from the top, and checking if the solution is prize-eligible. As a rule of thumb - any MagicLeap code/weights -- such as SuperPoint and/or SuperGlue disqualifies a team from money due to license restriction. ", "votes": 1 } ] }, { "id": 2289878, "postDate": "2023-06-06T11:59:22.493Z", "content": "@eduardtrulls @oldufo What about licenses for data used to train the models? For example, are models trained on scannet automatically treated as non-commercial, as scannet has a non-commercial license? For example loftr-indoor or dkm-indoor both are trained with scannet.
", "rawMarkdown": "@eduardtrulls @oldufo What about licenses for data used to train the models? For example, are models trained on scannet automatically treated as non-commercial, as scannet has a non-commercial license? For example loftr-indoor or dkm-indoor both are trained with scannet.", "replies": [ { "id": 2289984, "postDate": "2023-06-06T13:33:00.137Z", "content": "There are no indoor images in the competition, so using indoor models would not likely give you a top place.
", "rawMarkdown": "There are no indoor images in the competition, so using indoor models would not likely give you a top place. ", "replies": [ { "id": 2289992, "postDate": "2023-06-06T13:35:55.657Z", "content": "Sure, but e.g. the bike scene is very similar to typical indoor scenes. Indoor models can outperform there.
\nIn any case, the license issue needs clarification.
", "rawMarkdown": "Sure, but e.g. the bike scene is very similar to typical indoor scenes. Indoor models can outperform there.\n\nIn any case, the license issue needs clarification." }, { "id": 2292308, "postDate": "2023-06-08T07:57:14.153Z", "content": "What about the pre-trained models used for getting the most similar images in the function get_image_pairs_shortlist? As far as I am aware, ImageNet does not have a commercial license. Is it possible to perform fine-tuning on these models using publicly available data that are non-commercial?
", "rawMarkdown": "What about the pre-trained models used for getting the most similar images in the function get_image_pairs_shortlist? As far as I am aware, ImageNet does not have a commercial license. Is it possible to perform fine-tuning on these models using publicly available data that are non-commercial?" }, { "id": 2292469, "postDate": "2023-06-08T11:24:07.770Z", "content": "ImageNet is fine. ScanNet is not fine.
\nFor the others - there is a standard Kaggle practice, where you share the dataset and/or models and ask if that is OK to use them.
", "rawMarkdown": "ImageNet is fine. ScanNet is not fine. \n\nFor the others - there is a standard Kaggle practice, where you share the dataset and/or models and ask if that is OK to use them." } ] } ] }, { "id": 2219923, "postDate": "2023-04-13T01:58:33.953Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2219773, "postDate": "2023-04-12T21:16:21.230Z", "content": "thank you for the useful information
", "rawMarkdown": "thank you for the useful information", "votes": 1 } ], "comments": [ { "id": 2218649, "author_name": "Dieter", "author_url": "", "post_date": "2023-04-11T21:59:16.427000", "content": "\n\nUnfortunately this made a few top solutions unable to claim prizes.
\n
Thats quite an understatement. We were 12th place, yet got prize of 2nd place. Which means 10 teams before us were not eligible for the prize. Or if you see it differently, we followed the rules and got the 2nd best prize eligible solution, yet not even got a gold medal…
", "votes": 10, "replies": [ { "id": 2218654, "author_name": "old-ufo", "author_url": "", "post_date": "2023-04-11T22:09:46.303000", "content": "Well, the rules allowed non-open submission \"for glory only\", so 10 teams above you also followed rules. Which, arguable, were not ideal, but that's life. However we don't expect SG submissions to dominate this year.
", "votes": 0, "replies": [] }, { "id": 2223640, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-04-16T13:33:48.443000", "content": "True :)
\n2nd but without gold 😂 I remember.
If non-commercial code leads to higher ranks but doesn't allow to claim money prize, does it mean that each team have to decide what they are competing for? I'm speaking mostly about situation from last year's challenge, but the scenario could be the same this year. Could you please clarify one moment, will you judge only one submission? What if I submit one with higher score but with non-commercial code and second one with lower score but allowed license. Can I get gold based on first submit and prize based on second together? Thanks!
", "votes": 1, "replies": [ { "id": 2234034, "author_name": "Sohier Dane", "author_url": "", "post_date": "2023-04-24T19:44:14.460000", "content": "We're only going to review the highest scoring submission from each team.
", "votes": 1, "replies": [] } ] }, { "id": 2219704, "author_name": "Man of the year", "author_url": "", "post_date": "2023-04-12T19:53:58.820000", "content": "@eduardtrulls @oldufo hello, I didn't fully understand requirements to be invited to the CVPR workshop. Should we win a gold model, should we be a price contender?
", "votes": 1, "replies": [ { "id": 2219715, "author_name": "old-ufo", "author_url": "", "post_date": "2023-04-12T20:04:16.030000", "content": "The result should be good enough, and the solution should be interesting enough. Usually being in top-10 is good enough, but this doesn't mean that we automatically invite all top-10 teams, or that 20th place cannot be invited.
\nUnlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition.
", "votes": 3, "replies": [ { "id": 2219775, "author_name": "Man of the year", "author_url": "", "post_date": "2023-04-12T21:19:23.633000", "content": "Should the solution be among the chosen (in kaggle for private scoring) to be considered for the workshop?
", "votes": 0, "replies": [] }, { "id": 2219779, "author_name": "old-ufo", "author_url": "", "post_date": "2023-04-12T21:23:54.670000", "content": "I can only repeat why I have already said:
\n\n", "votes": 0, "replies": [] } ] } ] }, { "id": 2321576, "author_name": "Socratis Gkelios", "author_url": "", "post_date": "2023-06-28T17:16:11.733000", "content": "Unlike the prizes, which has clear criterion - private leaderboard, the CVPR Workshop invites are more subjective, and that is by design. We will decide that within Workshop organizers team the end of the competition.
\n
Is there any update on the prize winners?
", "votes": 0, "replies": [] }, { "id": 2300925, "author_name": "glowworm", "author_url": "", "post_date": "2023-06-13T14:16:01.790000", "content": "What are the five teams that got the bonus this year?
", "votes": 0, "replies": [ { "id": 2301111, "author_name": "old-ufo", "author_url": "", "post_date": "2023-06-13T16:15:03.760000", "content": "After the result verification by Kaggle team, they and we will reach out starting from the top, and checking if the solution is prize-eligible. As a rule of thumb - any MagicLeap code/weights -- such as SuperPoint and/or SuperGlue disqualifies a team from money due to license restriction.
", "votes": 1, "replies": [] } ] }, { "id": 2289878, "author_name": "Johan Edstedt", "author_url": "", "post_date": "2023-06-06T11:59:22.493000", "content": "@eduardtrulls @oldufo What about licenses for data used to train the models? For example, are models trained on scannet automatically treated as non-commercial, as scannet has a non-commercial license? For example loftr-indoor or dkm-indoor both are trained with scannet.
", "votes": 0, "replies": [ { "id": 2289984, "author_name": "old-ufo", "author_url": "", "post_date": "2023-06-06T13:33:00.137000", "content": "There are no indoor images in the competition, so using indoor models would not likely give you a top place.
", "votes": 0, "replies": [ { "id": 2289992, "author_name": "Johan Edstedt", "author_url": "", "post_date": "2023-06-06T13:35:55.657000", "content": "Sure, but e.g. the bike scene is very similar to typical indoor scenes. Indoor models can outperform there.
\nIn any case, the license issue needs clarification.
", "votes": 0, "replies": [] }, { "id": 2292308, "author_name": "Socratis Gkelios", "author_url": "", "post_date": "2023-06-08T07:57:14.153000", "content": "What about the pre-trained models used for getting the most similar images in the function get_image_pairs_shortlist? As far as I am aware, ImageNet does not have a commercial license. Is it possible to perform fine-tuning on these models using publicly available data that are non-commercial?
", "votes": 0, "replies": [] }, { "id": 2292469, "author_name": "old-ufo", "author_url": "", "post_date": "2023-06-08T11:24:07.770000", "content": "ImageNet is fine. ScanNet is not fine.
\nFor the others - there is a standard Kaggle practice, where you share the dataset and/or models and ask if that is OK to use them.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2219923, "author_name": "", "author_url": "", "post_date": "2023-04-13T01:58:33.953000", "content": "", "votes": 0, "replies": [] }, { "id": 2219773, "author_name": "Arciel Baez", "author_url": "", "post_date": "2023-04-12T21:16:21.230000", "content": "thank you for the useful information
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2218588": "We noticed in [last year's challenge](https://www.kaggle.com/competitions/image-matching-challenge-2022/overview) that some participants relied on code with non-commercial licenses for their solutions: specifically, the official implementations of [SuperPoint](https://github.com/magicleap/SuperPointPretrainedNetwork) and [SuperGlue](https://github.com/magicleap/SuperGluePretrainedNetwork). Unfortunately this made a few top solutions unable to claim prizes.\n\nWhile you are ultimately responsible to ensure that your submission complies with the [competition rules](https://www.kaggle.com/competitions/image-matching-challenge-2023/rules), we will provide guidance if possible. For example, there are other implementations of [SuperPoint](https://github.com/rpautrat/SuperPoint). We also hope to be able to point you to a new implementation of SuperGlue soon.)", "2218649": ">Unfortunately this made a few top solutions unable to claim prizes.\n\nThats quite an understatement. We were 12th place, yet got prize of 2nd place. Which means 10 teams before us were not eligible for the prize. Or if you see it differently, we followed the rules and got the 2nd best prize eligible solution, yet not even got a gold medal...", "2234001": "If non-commercial code leads to higher ranks but doesn't allow to claim money prize, does it mean that each team have to decide what they are competing for? I'm speaking mostly about situation from last year's challenge, but the scenario could be the same this year. Could you please clarify one moment, will you judge only one submission? What if I submit one with higher score but with non-commercial code and second one with lower score but allowed license. Can I get gold based on first submit and prize based on second together? Thanks!", "2219704": "@eduardtrulls @oldufo hello, I didn't fully understand requirements to be invited to the CVPR workshop. Should we win a gold model, should we be a price contender? ", "2321576": "Is there any update on the prize winners?", "2300925": "What are the five teams that got the bonus this year?", "2289878": "@eduardtrulls @oldufo What about licenses for data used to train the models? For example, are models trained on scannet automatically treated as non-commercial, as scannet has a non-commercial license? For example loftr-indoor or dkm-indoor both are trained with scannet.", "2219923": "", "2219773": "thank you for the useful information" } }