Hi everyone,
\nI would like to welcome you to the 2023 edition of the Image Matching Challenge! This is the second edition we held at Kaggle, which was quite succesful last year: you can check out the recap thread if you're curious. This year's challenge is a bit different though: instead of matching two images at a time, you'll be building 3D reconstructions from larger image sets!
\nThe competition is a bit different than most of the problems usually presented here, so we would like to provide you with a few links to get you started:
\nAnd some example notebooks:
\nWe plan to add more resources in the following days, so you may want to follow this thread.
\nBest of luck!
\n~ The organizers
", "messages": [ { "id": 2218421, "postDate": "2023-04-11T16:40:29.980Z", "content": "Hi everyone,
\nI would like to welcome you to the 2023 edition of the Image Matching Challenge! This is the second edition we held at Kaggle, which was quite succesful last year: you can check out the recap thread if you're curious. This year's challenge is a bit different though: instead of matching two images at a time, you'll be building 3D reconstructions from larger image sets!
\nThe competition is a bit different than most of the problems usually presented here, so we would like to provide you with a few links to get you started:
\nAnd some example notebooks:
\nWe plan to add more resources in the following days, so you may want to follow this thread.
\nBest of luck!
\n~ The organizers
", "rawMarkdown": "Hi everyone,\n\nI would like to welcome you to the 2023 edition of the Image Matching Challenge! This is the second edition we held at Kaggle, which was quite succesful last year: you can check out the [recap thread](https://www.kaggle.com/competitions/image-matching-challenge-2022/discussion/329650) if you're curious. This year's challenge is a bit different though: instead of matching two images at a time, you'll be building 3D reconstructions from larger image sets!\n\nThe competition is a bit different than most of the problems usually presented here, so we would like to provide you with a few links to get you started:\n\n* [Link to the workshop page](https://image-matching-workshop.github.io).\n* [Link to the previous version of the challenge](https://www.kaggle.com/competitions/image-matching-challenge-2022/overview).\n* [IJCV paper](https://arxiv.org/abs/2003.01587): a paper we published on this problem/data which will give you some context.\n\nAnd some example notebooks:\n\n* [Loading and inspecting the training data](https://www.kaggle.com/code/eduardtrulls/imc2023-training-data/notebook).\n * This is the easiest way to get started. It contains examples to reconstruct a 3D scene and visualize it.\n* [Creating a simple submission on GPU](https://www.kaggle.com/code/eduardtrulls/imc-2023-submission-example), using LoFTR or DISK through Kaggle models, and KeyNetAffNetHardNet through Kornia.\n * This notebook contains more advanced auxiliary functions which may help you interact with Colmap, the 3D reconstruction framework bundled with the competition.\n* [Evaluating a csv file on the training data with a single function call](https://www.kaggle.com/eduardtrulls/imc2023-evaluation/).\n * With this snippet you can evaluate your solution on the training data.\n\nWe plan to add more resources in the following days, so you may want to follow this thread.\n\nBest of luck!\n\n~ The organizers", "votes": 35 }, { "id": 2236226, "postDate": "2023-04-26T16:26:30.500Z", "content": "Last year we had situation where a lot of new accounts were created during competition (with the same result). I would like to ask host to watching on this activity - as far as I uderstand according to Kaggle rules it is now allowed to have multi account. Thank you!
", "rawMarkdown": "Last year we had situation where a lot of new accounts were created during competition (with the same result). I would like to ask host to watching on this activity - as far as I uderstand according to Kaggle rules it is now allowed to have multi account. Thank you!", "votes": 5, "replies": [ { "id": 2236234, "postDate": "2023-04-26T16:33:20.737Z", "content": "Thanks @remekkinas! Just to clarify - for the rules, we did not make any changes and users cannot sign up to Kaggle from multiple accounts:
\n\n", "rawMarkdown": "Thanks @remekkinas! Just to clarify - for the rules, we did not make any changes and users cannot sign up to Kaggle from multiple accounts:\n\n>**One account per participant**\nYou cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.", "replies": [ { "id": 2236307, "postDate": "2023-04-26T18:03:22.197Z", "content": "One account per participant
\n
\n You cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.
I think that last year we had many \"technical\" accounts - temporary made (with temporary made email accounts) for additional submission (it is better to have 10 submission for two accounts then 5 for one :)). I am writing this post on beggining of this competition. Last year we started to write about this on the end of competition when a lot of TOP20 submissions were new accounts (maybe this is OK … I do not know but I remeber people were surprised how much new accounts were created in TOP LB submission).
", "rawMarkdown": "I think that last year we had many \"technical\" accounts - temporary made (with temporary made email accounts) for additional submission (it is better to have 10 submission for two accounts then 5 for one :)). I am writing this post on beggining of this competition. Last year we started to write about this on the end of competition when a lot of TOP20 submissions were new accounts (maybe this is OK ... I do not know but I remeber people were surprised how much new accounts were created in TOP LB submission). ", "votes": 4 }, { "id": 2236504, "postDate": "2023-04-26T23:13:34.963Z", "content": "Hi Remek,
\nMultiple accounts, even if \"technical\" or temporary accounts, are still in violation of both competition and site rules. Kaggle has the right to remove all accounts associated with multiple accounts from the competition, and even potentially ban accounts from the platform altogether for violations. Removals of accounts typically occur at the end of the competition.
", "rawMarkdown": "Hi Remek,\n\nMultiple accounts, even if \"technical\" or temporary accounts, are still in violation of both competition and site rules. Kaggle has the right to remove all accounts associated with multiple accounts from the competition, and even potentially ban accounts from the platform altogether for violations. Removals of accounts typically occur at the end of the competition. ", "votes": 4 }, { "id": 2237014, "postDate": "2023-04-27T10:04:10.963Z", "content": "Thank you for your answer. 👍
", "rawMarkdown": "Thank you for your answer. 👍" }, { "id": 2283484, "postDate": "2023-06-01T09:35:58.700Z", "content": "Hi @ashleychow, today an account ranks top20(solo) is deleted, I think it means one of the teams rank top20 have multiple accounts.
", "rawMarkdown": "Hi @ashleychow, today an account ranks top20(solo) is deleted, I think it means one of the teams rank top20 have multiple accounts.", "votes": 2 } ] }, { "id": 2283480, "postDate": "2023-06-01T09:32:15.757Z", "content": "An account ranks top 20(solo) is deleted, it seems that one of the teams in top20 owns that account 👀.
", "rawMarkdown": "An account ranks top 20(solo) is deleted, it seems that one of the teams in top20 owns that account 👀.", "votes": 5, "replies": [ { "id": 2283513, "postDate": "2023-06-01T10:02:17.283Z", "content": "Magic in progress :)
", "rawMarkdown": "Magic in progress :)", "votes": 2 }, { "id": 2285152, "postDate": "2023-06-02T13:37:05.793Z", "content": "This is not the only deleted account, I can see at least 4 for now. The geese in gold zone are sus for sure.
", "rawMarkdown": "This is not the only deleted account, I can see at least 4 for now. The geese in gold zone are sus for sure.", "votes": 2 } ] } ] }, { "id": 2220777, "postDate": "2023-04-13T17:08:07.077Z", "content": "Does test dataset contain new scenes?
", "rawMarkdown": "Does test dataset contain new scenes?", "votes": 1, "replies": [ { "id": 2220782, "postDate": "2023-04-13T17:18:41.700Z", "content": "Yes, there are new scenes wrt the previous edition of the IMC and these new scenes are those this year competition focuses on.
\nYou can check here: https://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis
Thank you for your reply. but my question is whether the test dataset contains any scenes that are not included in the training dataset. Apologies if this information is already provided elsewhere.
", "rawMarkdown": "Thank you for your reply. but my question is whether the test dataset contains any scenes that are not included in the training dataset. Apologies if this information is already provided elsewhere." }, { "id": 2220812, "postDate": "2023-04-13T17:36:00.697Z", "content": "Yes, training data provided are disjoint from those used to get the public or private leaderboards.
", "rawMarkdown": "Yes, training data provided are disjoint from those used to get the public or private leaderboards." }, { "id": 2220828, "postDate": "2023-04-13T17:43:52.323Z", "content": "Sorry, but that would be incredibly stupid for us to mix test set with training :)
", "rawMarkdown": "Sorry, but that would be incredibly stupid for us to mix test set with training :)", "votes": 1 }, { "id": 2220831, "postDate": "2023-04-13T17:44:46.657Z", "content": "I got it, thanks :)
", "rawMarkdown": "I got it, thanks :)" } ] } ] }, { "id": 2277858, "postDate": "2023-05-28T05:52:45.477Z", "content": "really excited
", "rawMarkdown": "really excited" }, { "id": 2269395, "postDate": "2023-05-22T12:55:26.717Z", "content": "really helpful
", "rawMarkdown": "really helpful" }, { "id": 2250011, "postDate": "2023-05-08T09:24:44.817Z", "content": "How to get the the score without submitting the code( checking the score)?
", "rawMarkdown": "How to get the the score without submitting the code( checking the score)?", "replies": [ { "id": 2250069, "postDate": "2023-05-08T10:18:54.340Z", "content": "You need to make a submission to get the score.
", "rawMarkdown": "You need to make a submission to get the score.", "replies": [ { "id": 2250082, "postDate": "2023-05-08T10:28:54.777Z", "content": "I think we can use the evaluation notebook for that isn't it?
", "rawMarkdown": "I think we can use the evaluation notebook for that isn't it?" } ] } ] }, { "id": 2243877, "postDate": "2023-05-03T08:48:26.857Z", "content": "Test dataset contain new scenes is this Necessary?
", "rawMarkdown": "Test dataset contain new scenes is this Necessary?", "replies": [ { "id": 2268085, "postDate": "2023-05-21T12:49:18.393Z", "content": "Do you get the answer?I want to know about it too.
", "rawMarkdown": "Do you get the answer?I want to know about it too." }, { "id": 2268892, "postDate": "2023-05-22T05:15:45.890Z", "content": "Could you share the answer?
", "rawMarkdown": "Could you share the answer?", "replies": [ { "id": 2268997, "postDate": "2023-05-22T06:43:16.703Z", "content": "Yes, the test dataset contains different scenes than the training dataset.
", "rawMarkdown": "Yes, the test dataset contains different scenes than the training dataset." }, { "id": 2273563, "postDate": "2023-05-25T08:26:41.887Z", "content": "What I understand is that the scenes in the test set and the training set are the same, but the photos taken from different camera pose. Is my understanding wrong?
", "rawMarkdown": "What I understand is that the scenes in the test set and the training set are the same, but the photos taken from different camera pose. Is my understanding wrong?" }, { "id": 2273693, "postDate": "2023-05-25T10:39:33.830Z", "content": "Yes, your understanding is wrong. Test and training set are taken in completely different locations.
", "rawMarkdown": "Yes, your understanding is wrong. Test and training set are taken in completely different locations." }, { "id": 2274405, "postDate": "2023-05-26T00:53:45.817Z", "content": "got it, thanks
", "rawMarkdown": "got it, thanks" } ] } ] }, { "id": 2225665, "postDate": "2023-04-18T10:40:50.373Z", "content": "Thank you for providing the new IMC competition! I have participated for two years. I noticed that this year's competition focuses on 3D reconstruction, but the evaluation metrics are still prepared for pose estimation. Why not include the evaluation metrics from MVS into this competition? It would be more reasonable to submit a reconstructed point cloud model.
", "rawMarkdown": "Thank you for providing the new IMC competition! I have participated for two years. I noticed that this year's competition focuses on 3D reconstruction, but the evaluation metrics are still prepared for pose estimation. Why not include the evaluation metrics from MVS into this competition? It would be more reasonable to submit a reconstructed point cloud model.", "replies": [ { "id": 2225684, "postDate": "2023-04-18T10:50:53.773Z", "content": "There are many great benchmarks for MVS, such as T&T. There are two reasons we are not focusing on SfM part.
\n1) SfM is far from solved. The data we provide is hard enough to make vanilla Colmap fail way before we get to the MVS stage.
\nWhat is the value of point cloud if one cannot register camera correctly?
2) Reliable GT. Making sure that our camera poses are good enough is hard and costly already. Making sure that we have super-precise GT for the point cloud is MUCH harder and expensive. You can check LaMAR benchmark to get an idea of such process.
", "rawMarkdown": "There are many great benchmarks for MVS, such as [T&T](https://www.tanksandtemples.org). There are two reasons we are not focusing on SfM part.\n\n1) SfM is far from solved. The data we provide is hard enough to make vanilla Colmap fail way before we get to the MVS stage. \nWhat is the value of point cloud if one cannot register camera correctly? \n\n2) Reliable GT. Making sure that our camera poses are good enough is hard and costly already. Making sure that we have super-precise GT for the point cloud is MUCH harder and expensive. You can check [LaMAR benchmark](https://lamar.ethz.ch) to get an idea of such process.", "votes": 4, "replies": [ { "id": 2226552, "postDate": "2023-04-19T03:15:49.620Z", "content": "Thanks for your sincere reply!
", "rawMarkdown": "Thanks for your sincere reply!" } ] } ] }, { "id": 2223417, "postDate": "2023-04-16T08:35:56.970Z", "content": "@fabiobellavia @oldufo Thanks for the new IMC!
\nit seems that some GT data is missing in train folder. For some scenes in the 'phototourism' folder sfm data is not provided e.g. \"temple_nara_japan\". And, scenes listed in 'phototourism' folder are not given in train_labels.csv. Could you check it?
The train_labels.csv is only provided for the three new datasets, and that's what we expect participants will evaluate on. The phototourism data is the same as last year, with the new sfm folders. You can certainly use the phototourism dataset for evaluation if you want(*), but note that it may contain incorrectly posed images, as we have not verified all of them by hand.
(*) It looks like the calibration.csv files (which contain R and T) didn't survive the move to the 2023 dataset, not sure why. It's probably easiest to just pull the 2022 data if you want to use it.
As for the temple_nara_japan sfm folder missing: probably a mistake on my part. The sfm model is only used for inspection and updating the dataset is quite involved, so we'll probably leave it as-is.
", "rawMarkdown": "The `train_labels.csv` is only provided for the three new datasets, and that's what we expect participants will evaluate on. The phototourism data is the same as last year, with the new sfm folders. You can certainly use the phototourism dataset for evaluation if you want(*), but note that it may contain incorrectly posed images, as we have not verified all of them by hand. \n\n(*) It looks like the `calibration.csv` files (which contain R and T) didn't survive the move to the 2023 dataset, not sure why. It's probably easiest to just pull the 2022 data if you want to use it.\n\nAs for the temple_nara_japan sfm folder missing: probably a mistake on my part. The sfm model is only used for inspection and updating the dataset is quite involved, so we'll probably leave it as-is.", "votes": 1 } ] }, { "id": 2259024, "postDate": "2023-05-14T16:25:31.247Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2219926, "postDate": "2023-04-13T02:06:23.873Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 2267791, "postDate": "2023-05-21T08:34:05.183Z", "content": "thanks，dude
", "rawMarkdown": "thanks，dude" }, { "id": 2265092, "postDate": "2023-05-19T01:21:03.247Z", "content": "really helpful，thanks
", "rawMarkdown": "really helpful，thanks\n" } ], "comments": [ { "id": 2236226, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-04-26T16:26:30.500000", "content": "Last year we had situation where a lot of new accounts were created during competition (with the same result). I would like to ask host to watching on this activity - as far as I uderstand according to Kaggle rules it is now allowed to have multi account. Thank you!
", "votes": 5, "replies": [ { "id": 2236234, "author_name": "Ashley Chow", "author_url": "", "post_date": "2023-04-26T16:33:20.737000", "content": "Thanks @remekkinas! Just to clarify - for the rules, we did not make any changes and users cannot sign up to Kaggle from multiple accounts:
\n\n", "votes": 0, "replies": [ { "id": 2236307, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-04-26T18:03:22.197000", "content": "One account per participant
\n
\n You cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.
I think that last year we had many \"technical\" accounts - temporary made (with temporary made email accounts) for additional submission (it is better to have 10 submission for two accounts then 5 for one :)). I am writing this post on beggining of this competition. Last year we started to write about this on the end of competition when a lot of TOP20 submissions were new accounts (maybe this is OK … I do not know but I remeber people were surprised how much new accounts were created in TOP LB submission).
", "votes": 4, "replies": [] }, { "id": 2236504, "author_name": "Addison Howard", "author_url": "", "post_date": "2023-04-26T23:13:34.963000", "content": "Hi Remek,
\nMultiple accounts, even if \"technical\" or temporary accounts, are still in violation of both competition and site rules. Kaggle has the right to remove all accounts associated with multiple accounts from the competition, and even potentially ban accounts from the platform altogether for violations. Removals of accounts typically occur at the end of the competition.
", "votes": 4, "replies": [] }, { "id": 2237014, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-04-27T10:04:10.963000", "content": "Thank you for your answer. 👍
", "votes": 0, "replies": [] }, { "id": 2283484, "author_name": "ForcewithMe", "author_url": "", "post_date": "2023-06-01T09:35:58.700000", "content": "Hi @ashleychow, today an account ranks top20(solo) is deleted, I think it means one of the teams rank top20 have multiple accounts.
", "votes": 2, "replies": [] } ] }, { "id": 2283480, "author_name": "ForcewithMe", "author_url": "", "post_date": "2023-06-01T09:32:15.757000", "content": "An account ranks top 20(solo) is deleted, it seems that one of the teams in top20 owns that account 👀.
", "votes": 5, "replies": [ { "id": 2283513, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-06-01T10:02:17.283000", "content": "Magic in progress :)
", "votes": 2, "replies": [] }, { "id": 2285152, "author_name": "Vladislav Ostankovich", "author_url": "", "post_date": "2023-06-02T13:37:05.793000", "content": "This is not the only deleted account, I can see at least 4 for now. The geese in gold zone are sus for sure.
", "votes": 2, "replies": [] } ] } ] }, { "id": 2220777, "author_name": "T.N", "author_url": "", "post_date": "2023-04-13T17:08:07.077000", "content": "Does test dataset contain new scenes?
", "votes": 1, "replies": [ { "id": 2220782, "author_name": "WastedCode", "author_url": "", "post_date": "2023-04-13T17:18:41.700000", "content": "Yes, there are new scenes wrt the previous edition of the IMC and these new scenes are those this year competition focuses on.
\nYou can check here: https://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis
Thank you for your reply. but my question is whether the test dataset contains any scenes that are not included in the training dataset. Apologies if this information is already provided elsewhere.
", "votes": 0, "replies": [] }, { "id": 2220812, "author_name": "WastedCode", "author_url": "", "post_date": "2023-04-13T17:36:00.697000", "content": "Yes, training data provided are disjoint from those used to get the public or private leaderboards.
", "votes": 0, "replies": [] }, { "id": 2220828, "author_name": "old-ufo", "author_url": "", "post_date": "2023-04-13T17:43:52.323000", "content": "Sorry, but that would be incredibly stupid for us to mix test set with training :)
", "votes": 1, "replies": [] }, { "id": 2220831, "author_name": "T.N", "author_url": "", "post_date": "2023-04-13T17:44:46.657000", "content": "I got it, thanks :)
", "votes": 0, "replies": [] } ] } ] }, { "id": 2277858, "author_name": "Alienssssss", "author_url": "", "post_date": "2023-05-28T05:52:45.477000", "content": "really excited
", "votes": 0, "replies": [] }, { "id": 2269395, "author_name": "WUHU", "author_url": "", "post_date": "2023-05-22T12:55:26.717000", "content": "really helpful
", "votes": 0, "replies": [] }, { "id": 2250011, "author_name": "UoM_200296M", "author_url": "", "post_date": "2023-05-08T09:24:44.817000", "content": "How to get the the score without submitting the code( checking the score)?
", "votes": 0, "replies": [ { "id": 2250069, "author_name": "Eduard Trulls", "author_url": "", "post_date": "2023-05-08T10:18:54.340000", "content": "You need to make a submission to get the score.
", "votes": 0, "replies": [ { "id": 2250082, "author_name": "UoM_200296M", "author_url": "", "post_date": "2023-05-08T10:28:54.777000", "content": "I think we can use the evaluation notebook for that isn't it?
", "votes": 0, "replies": [] } ] } ] }, { "id": 2243877, "author_name": "Mohammed Naqeeb", "author_url": "", "post_date": "2023-05-03T08:48:26.857000", "content": "Test dataset contain new scenes is this Necessary?
", "votes": 0, "replies": [ { "id": 2268085, "author_name": "Tarnished Q", "author_url": "", "post_date": "2023-05-21T12:49:18.393000", "content": "Do you get the answer?I want to know about it too.
", "votes": 0, "replies": [] }, { "id": 2268892, "author_name": "JavaScript996", "author_url": "", "post_date": "2023-05-22T05:15:45.890000", "content": "Could you share the answer?
", "votes": 0, "replies": [ { "id": 2268997, "author_name": "Eduard Trulls", "author_url": "", "post_date": "2023-05-22T06:43:16.703000", "content": "Yes, the test dataset contains different scenes than the training dataset.
", "votes": 0, "replies": [] }, { "id": 2273563, "author_name": "LiBoKun", "author_url": "", "post_date": "2023-05-25T08:26:41.887000", "content": "What I understand is that the scenes in the test set and the training set are the same, but the photos taken from different camera pose. Is my understanding wrong?
", "votes": 0, "replies": [] }, { "id": 2273693, "author_name": "old-ufo", "author_url": "", "post_date": "2023-05-25T10:39:33.830000", "content": "Yes, your understanding is wrong. Test and training set are taken in completely different locations.
", "votes": 0, "replies": [] }, { "id": 2274405, "author_name": "LiBoKun", "author_url": "", "post_date": "2023-05-26T00:53:45.817000", "content": "got it, thanks
", "votes": 0, "replies": [] } ] } ] }, { "id": 2225665, "author_name": "Weiyue Zhao", "author_url": "", "post_date": "2023-04-18T10:40:50.373000", "content": "Thank you for providing the new IMC competition! I have participated for two years. I noticed that this year's competition focuses on 3D reconstruction, but the evaluation metrics are still prepared for pose estimation. Why not include the evaluation metrics from MVS into this competition? It would be more reasonable to submit a reconstructed point cloud model.
", "votes": 0, "replies": [ { "id": 2225684, "author_name": "old-ufo", "author_url": "", "post_date": "2023-04-18T10:50:53.773000", "content": "There are many great benchmarks for MVS, such as T&T. There are two reasons we are not focusing on SfM part.
\n1) SfM is far from solved. The data we provide is hard enough to make vanilla Colmap fail way before we get to the MVS stage.
\nWhat is the value of point cloud if one cannot register camera correctly?
2) Reliable GT. Making sure that our camera poses are good enough is hard and costly already. Making sure that we have super-precise GT for the point cloud is MUCH harder and expensive. You can check LaMAR benchmark to get an idea of such process.
", "votes": 4, "replies": [ { "id": 2226552, "author_name": "Weiyue Zhao", "author_url": "", "post_date": "2023-04-19T03:15:49.620000", "content": "Thanks for your sincere reply!
", "votes": 0, "replies": [] } ] } ] }, { "id": 2223417, "author_name": "Igor Lashkov", "author_url": "", "post_date": "2023-04-16T08:35:56.970000", "content": "@fabiobellavia @oldufo Thanks for the new IMC!
\nit seems that some GT data is missing in train folder. For some scenes in the 'phototourism' folder sfm data is not provided e.g. \"temple_nara_japan\". And, scenes listed in 'phototourism' folder are not given in train_labels.csv. Could you check it?
The train_labels.csv is only provided for the three new datasets, and that's what we expect participants will evaluate on. The phototourism data is the same as last year, with the new sfm folders. You can certainly use the phototourism dataset for evaluation if you want(*), but note that it may contain incorrectly posed images, as we have not verified all of them by hand.
(*) It looks like the calibration.csv files (which contain R and T) didn't survive the move to the 2023 dataset, not sure why. It's probably easiest to just pull the 2022 data if you want to use it.
As for the temple_nara_japan sfm folder missing: probably a mistake on my part. The sfm model is only used for inspection and updating the dataset is quite involved, so we'll probably leave it as-is.
", "votes": 1, "replies": [] } ] }, { "id": 2259024, "author_name": "", "author_url": "", "post_date": "2023-05-14T16:25:31.247000", "content": "", "votes": 0, "replies": [] }, { "id": 2219926, "author_name": "", "author_url": "", "post_date": "2023-04-13T02:06:23.873000", "content": "", "votes": 1, "replies": [] }, { "id": 2267791, "author_name": "Gzwwhhh", "author_url": "", "post_date": "2023-05-21T08:34:05.183000", "content": "thanks，dude
", "votes": 0, "replies": [] }, { "id": 2265092, "author_name": "吕舟", "author_url": "", "post_date": "2023-05-19T01:21:03.247000", "content": "really helpful，thanks
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2218421": "Hi everyone,\n\nI would like to welcome you to the 2023 edition of the Image Matching Challenge! This is the second edition we held at Kaggle, which was quite succesful last year: you can check out the [recap thread](https://www.kaggle.com/competitions/image-matching-challenge-2022/discussion/329650) if you're curious. This year's challenge is a bit different though: instead of matching two images at a time, you'll be building 3D reconstructions from larger image sets!\n\nThe competition is a bit different than most of the problems usually presented here, so we would like to provide you with a few links to get you started:\n\n* [Link to the workshop page](https://image-matching-workshop.github.io).\n* [Link to the previous version of the challenge](https://www.kaggle.com/competitions/image-matching-challenge-2022/overview).\n* [IJCV paper](https://arxiv.org/abs/2003.01587): a paper we published on this problem/data which will give you some context.\n\nAnd some example notebooks:\n\n* [Loading and inspecting the training data](https://www.kaggle.com/code/eduardtrulls/imc2023-training-data/notebook).\n * This is the easiest way to get started. It contains examples to reconstruct a 3D scene and visualize it.\n* [Creating a simple submission on GPU](https://www.kaggle.com/code/eduardtrulls/imc-2023-submission-example), using LoFTR or DISK through Kaggle models, and KeyNetAffNetHardNet through Kornia.\n * This notebook contains more advanced auxiliary functions which may help you interact with Colmap, the 3D reconstruction framework bundled with the competition.\n* [Evaluating a csv file on the training data with a single function call](https://www.kaggle.com/eduardtrulls/imc2023-evaluation/).\n * With this snippet you can evaluate your solution on the training data.\n\nWe plan to add more resources in the following days, so you may want to follow this thread.\n\nBest of luck!\n\n~ The organizers", "2236226": "Last year we had situation where a lot of new accounts were created during competition (with the same result). I would like to ask host to watching on this activity - as far as I uderstand according to Kaggle rules it is now allowed to have multi account. Thank you!", "2220777": "Does test dataset contain new scenes?", "2277858": "really excited", "2269395": "really helpful", "2250011": "How to get the the score without submitting the code( checking the score)?", "2243877": "Test dataset contain new scenes is this Necessary?", "2225665": "Thank you for providing the new IMC competition! I have participated for two years. I noticed that this year's competition focuses on 3D reconstruction, but the evaluation metrics are still prepared for pose estimation. Why not include the evaluation metrics from MVS into this competition? It would be more reasonable to submit a reconstructed point cloud model.", "2223417": "@fabiobellavia @oldufo Thanks for the new IMC! \nit seems that some GT data is missing in train folder. For some scenes in the 'phototourism' folder sfm data is not provided e.g. \"temple_nara_japan\". And, scenes listed in 'phototourism' folder are not given in train_labels.csv. Could you check it?", "2259024": "", "2219926": "", "2267791": "thanks，dude", "2265092": "really helpful，thanks\n" } }