{ "id": 486584, "title": "Welcome to the 2024 Image Matching Challenge!", "url": "/competitions/image-matching-challenge-2024/discussion/486584", "author_name": "old-ufo", "post_date": "2024-03-25T15:33:06.441000", "votes": 55, "comment_count": 29, "views": 0, "content": "

Hi everyone,

\n

I would like to welcome you to the 2024 edition of the Image Matching Challenge! This is the third edition we held at Kaggle. Both of the previous editions were quite successful: you can check out the recap thread if you're curious.

\n

This year's challenge we raise the difficulty level of the task even more. Same as a last year you'll be building 3D reconstructions from medium-sized image sets: up to 100 images.

\n

However, instead of dealing with a single nuisance factor (e.g. in plane rotation for the historical preservation data last year), you will be dealing with multiple of them in the same time. For example, the images are taken from different view points AND at different time of the day AND during a different season. Or, for example, the data is of nature origin (trees, plants, flowers), and contains repeated patterns.

\n

The 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

And some example notebooks:

\n\n

Important: you can also look at the winner notebooks at 2023 competition. However, they will not work \"out-of-the-box\" because the pycolmap version changed between the competitions.

\n

The main differences in interface are in:

\n\n

We plan to add more resources in the following days, so you may want to follow this thread.

\n

Prize-eligibility

\n\n

Best of luck!

\n

~ The organizers

", "messages": [ { "id": 2715623, "postDate": "2024-03-25T15:33:06.440Z", "content": "

Hi everyone,

\n

I would like to welcome you to the 2024 edition of the Image Matching Challenge! This is the third edition we held at Kaggle. Both of the previous editions were quite successful: you can check out the recap thread if you're curious.

\n

This year's challenge we raise the difficulty level of the task even more. Same as a last year you'll be building 3D reconstructions from medium-sized image sets: up to 100 images.

\n

However, instead of dealing with a single nuisance factor (e.g. in plane rotation for the historical preservation data last year), you will be dealing with multiple of them in the same time. For example, the images are taken from different view points AND at different time of the day AND during a different season. Or, for example, the data is of nature origin (trees, plants, flowers), and contains repeated patterns.

\n

The 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

And some example notebooks:

\n\n

Important: you can also look at the winner notebooks at 2023 competition. However, they will not work \"out-of-the-box\" because the pycolmap version changed between the competitions.

\n

The main differences in interface are in:

\n\n

We plan to add more resources in the following days, so you may want to follow this thread.

\n

Prize-eligibility

\n\n

Best of luck!

\n

~ The organizers

", "rawMarkdown": "Hi everyone,\n\nI would like to welcome you to the 2024 edition of the Image Matching Challenge! This is the third edition we held at Kaggle. Both of the previous editions were quite successful: you can check out the recap thread if you're curious. \n\nThis year's challenge we raise the difficulty level of the task even more. Same as a last year you'll be building 3D reconstructions from medium-sized image sets: up to 100 images. \n\nHowever, instead of dealing with a single nuisance factor (e.g. in plane rotation for the historical preservation data last year), you will be dealing with multiple of them in the same time. For example, the images are taken from different view points AND at different time of the day AND during a different season. Or, for example, the data is of nature origin (trees, plants, flowers), and contains repeated patterns.\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- Links to the previous versions of the challenge: [2023](https://www.kaggle.com/competitions/image-matching-challenge-2023), [2022](https://www.kaggle.com/competitions/image-matching-challenge-2022)\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- [Creating a simple submission on GPU](https://www.kaggle.com/code/oldufo/imc-2024-submission-example), using ALIKED local features and LightGlue matcher through Kaggle models, and DINOv2 for creating a shortlist. This notebook contains more advanced auxiliary functions and datasets which may help you interact with Colmap, the 3D reconstruction framework bundled with the competition.\n- **Updated** [Running the evaluation metric on training set](https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example)\n- **Updated** [Training 3D model visualization with Rerun.io](https://www.kaggle.com/code/oldufo/colmap-3d-visualization-with-rerun-io)\n- **Updated** [Deep Image Matching toolbox example](https://www.kaggle.com/code/lcmrll/dim-package-submission-example)\n\n**Important**: you can also look at the [winner notebooks at 2023](https://www.kaggle.com/competitions/image-matching-challenge-2023/leaderboard) competition. However, they will not work \"out-of-the-box\" because the pycolmap version changed between the competitions. \n\nThe main differences in interface are in:\n\n- Reconstruction option object is renamed: `mapper_options = pycolmap.IncrementalMapperOptions()` in old version vs current `mapper_options = pycolmap.IncrementalPipelineOptions()`;\n- The structure, which contains camera pose: old version uses `im.rotmat(), im.tvec` vs `im.cam_from_world.rotation.matrix(), im.cam_from_world.translation` in new version.\n\nWe plan to add more resources in the following days, so you may want to follow this thread.\n\n### Prize-eligibility\n\n- Any submission, using non-commercial licensed 3rd party code is not prize eligible. E.g. submissions using SuperGlue (or SuperPoint) are not prize-eligible. Use LightGlue instead, which is both open source, and provides better performance. \n- GPL licensed 3rd party code is OK. However, your own code should be Apache 2.0 licensed. \n\nBest of luck!\n\n~ The organizers", "votes": 54 }, { "id": 2716833, "postDate": "2024-03-26T08:27:00.163Z", "content": "

We have also added the metric code calculation here https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example

", "rawMarkdown": "We have also added the metric code calculation here https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example", "votes": 7 }, { "id": 2718037, "postDate": "2024-03-26T22:36:43.950Z", "content": "

We also added interactive visualizations of the 3D models with Rerun.io
\nYou need to clone and run the notebook to see the visualization.

", "rawMarkdown": "We also added interactive visualizations of the [3D models with Rerun.io](https://www.kaggle.com/code/oldufo/colmap-3d-visualization-with-rerun-io)\nYou need to clone and run the notebook to see the visualization. ", "votes": 3 }, { "id": 2808692, "postDate": "2024-05-12T11:01:23.887Z", "content": "

What a great competition!

", "rawMarkdown": "What a great competition!", "votes": 1 }, { "id": 2744809, "postDate": "2024-04-10T06:22:56.693Z", "content": "

What is the relationship of the \"scene\" and \"dataset\" fields in the submission CSV and the directory structure in train/ and test/? The data page of the competition says \"[train/test]/*/*/images\" with two asterisks, presumably representing dataset and scene.

\n

However, in the dataset it seems that the directory structure only contains one level. Is that dataset or scene?

\n
$ tree -d \n\n├── test\n│   └── church\n│       └── images\n└── train\n    ├── church\n    │   ├── images\n    │   └── sfm\n    ├── dioscuri\n    │   ├── images\n    │   └── sfm\n    ├── lizard\n    │   ├── images\n    │   └── sfm\n    ├── multi-temporal-temple-baalshamin\n    │   ├── images\n    │   └── sfm\n    ├── pond\n    │   ├── images\n    │   └── sfm\n    ├── transp_obj_glass_cup\n    │   ├── images\n    │   └── sfm\n    └── transp_obj_glass_cylinder\n        ├── images\n        └── sfm\n
", "rawMarkdown": "What is the relationship of the \"scene\" and \"dataset\" fields in the submission CSV and the directory structure in train/ and test/? The data page of the competition says \"[train/test]/\\*/\\*/images\" with two asterisks, presumably representing dataset and scene.\n\nHowever, in the dataset it seems that the directory structure only contains one level. Is that dataset or scene?\n\n```\n$ tree -d data\ndata\n├── test\n│ └── church\n│ └── images\n└── train\n ├── church\n │ ├── images\n │ └── sfm\n ├── dioscuri\n │ ├── images\n │ └── sfm\n ├── lizard\n │ ├── images\n │ └── sfm\n ├── multi-temporal-temple-baalshamin\n │ ├── images\n │ └── sfm\n ├── pond\n │ ├── images\n │ └── sfm\n ├── transp_obj_glass_cup\n │ ├── images\n │ └── sfm\n └── transp_obj_glass_cylinder\n ├── images\n └── sfm\n```", "votes": 1, "replies": [ { "id": 2744970, "postDate": "2024-04-10T08:54:56.617Z", "content": "

This is a leftover from the previous competition directory structure. This year you are correct that dataset is the same as the scene.

", "rawMarkdown": "This is a leftover from the previous competition directory structure. This year you are correct that dataset is the same as the scene.", "votes": 4 } ] }, { "id": 2747521, "postDate": "2024-04-11T23:56:16.087Z", "content": "

I really love competition like this, preparing to get in touch

", "rawMarkdown": "I really love competition like this, preparing to get in touch" }, { "id": 2728379, "postDate": "2024-04-02T07:57:23.693Z", "content": "

We have added Deep Image Matching toolbox example https://www.kaggle.com/code/lcmrll/dim-package-submission-example

", "rawMarkdown": "We have added Deep Image Matching toolbox example https://www.kaggle.com/code/lcmrll/dim-package-submission-example", "votes": 2 }, { "id": 2834572, "postDate": "2024-05-24T20:22:05.830Z", "content": "

nice this is greate one

", "rawMarkdown": "nice this is greate one" }, { "id": 2819531, "postDate": "2024-05-17T15:10:18.537Z", "content": "

What a great competition!

", "rawMarkdown": "What a great competition!" }, { "id": 2814018, "postDate": "2024-05-15T05:52:25.563Z", "content": "

This competition is really interesting!

", "rawMarkdown": "This competition is really interesting!" }, { "id": 2810451, "postDate": "2024-05-13T09:20:00.603Z", "content": "

Can we skip using Colmap and instead use other 3D reconstruction frameworks？

", "rawMarkdown": "Can we skip using Colmap and instead use other 3D reconstruction frameworks？", "replies": [ { "id": 2810519, "postDate": "2024-05-13T09:49:30.130Z", "content": "

Of course you can.

", "rawMarkdown": "Of course you can." } ] }, { "id": 2792014, "postDate": "2024-05-04T03:24:09.257Z", "content": "

Why did I run the code without task issues, the commit file was generated normally, I also checked the commit file: if there is an error in the number of rows or columns, a null value, the data type of the value is incorrect, or the commit value is inconsistent with the expected value. But in the end, it shows a scoring error, how should this be solved, does anyone have a similar situation?

", "rawMarkdown": "Why did I run the code without task issues, the commit file was generated normally, I also checked the commit file: if there is an error in the number of rows or columns, a null value, the data type of the value is incorrect, or the commit value is inconsistent with the expected value. But in the end, it shows a scoring error, how should this be solved, does anyone have a similar situation?" }, { "id": 2780303, "postDate": "2024-04-28T06:12:55.950Z", "content": "

I like this competition

", "rawMarkdown": "I like this competition" }, { "id": 2759965, "postDate": "2024-04-19T03:21:27.433Z", "content": "

sounds good

", "rawMarkdown": "sounds good" }, { "id": 2753943, "postDate": "2024-04-15T18:12:37.320Z", "content": "

Are the images for a particular dataset all collected with with the same camera (e.g. are the intrinsics fixed for a dataset?)

", "rawMarkdown": "Are the images for a particular dataset all collected with with the same camera (e.g. are the intrinsics fixed for a dataset?)", "replies": [ { "id": 2754059, "postDate": "2024-04-15T19:13:23.580Z", "content": "

This is a dangerous assumption to make.

", "rawMarkdown": "This is a dangerous assumption to make.", "replies": [ { "id": 2766277, "postDate": "2024-04-21T15:39:25.903Z", "content": "

It seems difficult / potentially ill posed to solve SFM when all images are captured from cameras with different intrinsics and we have no prior info on these parameters. Is the EXIF data generally an accurate enough initial value for SFM? I also would be surprised if there not at least some images that have exactly the same intrinsics. Does anyone have any more intuition on this?

", "rawMarkdown": "It seems difficult / potentially ill posed to solve SFM when all images are captured from cameras with different intrinsics and we have no prior info on these parameters. Is the EXIF data generally an accurate enough initial value for SFM? I also would be surprised if there not at least some images that have exactly the same intrinsics. Does anyone have any more intuition on this? ", "votes": 1 } ] } ] }, { "id": 2727667, "postDate": "2024-04-01T21:59:58.797Z", "content": "

Hello! I can't figure out if OpenCV library (open-source part) is allowed to be used in this competition. Does anyone know?

", "rawMarkdown": "Hello! I can't figure out if OpenCV library (open-source part) is allowed to be used in this competition. Does anyone know?", "replies": [ { "id": 2728378, "postDate": "2024-04-02T07:52:56.047Z", "content": "

Of course it is allowed. Literally all open source libraries are allowed (MIT, GPL, BSD3, Apache2, etc), unless they are non-commercial license.

", "rawMarkdown": "Of course it is allowed. Literally all open source libraries are allowed (MIT, GPL, BSD3, Apache2, etc), unless they are non-commercial license. ", "votes": 1 } ] }, { "id": 2724666, "postDate": "2024-03-31T03:17:32.437Z", "content": "

sounds good

", "rawMarkdown": "sounds good", "replies": [ { "id": 2770571, "postDate": "2024-04-23T22:44:32.663Z", "content": "

sounds good

", "rawMarkdown": "sounds good" } ] }, { "id": 2719273, "postDate": "2024-03-27T15:38:45.477Z", "content": "

Can anyone explain what is the rotation matrix and translation vector

", "rawMarkdown": "Can anyone explain what is the rotation matrix and translation vector ", "replies": [ { "id": 2719427, "postDate": "2024-03-27T17:44:52.023Z", "content": "

You can check here https://en.wikipedia.org/wiki/Camera_matrix

\n

Or here https://kornia.readthedocs.io/en/latest/geometry.camera.pinhole.html#pinhole-camera

", "rawMarkdown": "You can check here https://en.wikipedia.org/wiki/Camera_matrix\n\nOr here https://kornia.readthedocs.io/en/latest/geometry.camera.pinhole.html#pinhole-camera" } ] }, { "id": 2852495, "postDate": "2024-06-03T10:19:26.547Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2852491, "postDate": "2024-06-03T10:15:58.910Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2737811, "postDate": "2024-04-06T00:34:43.030Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 2813093, "postDate": "2024-05-14T15:15:51.083Z", "content": "

+1 any answer?

", "rawMarkdown": "+1 any answer?" } ] }, { "id": 2797606, "postDate": "2024-05-06T19:55:16.443Z", "content": "

Thank you to COMPETITION HOST!

", "rawMarkdown": "Thank you to COMPETITION HOST!" } ], "comments": [ { "id": 2716833, "author_name": "old-ufo", "author_url": "", "post_date": "2024-03-26T08:27:00.163000", "content": "

We have also added the metric code calculation here https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example

", "votes": 7, "replies": [] }, { "id": 2718037, "author_name": "old-ufo", "author_url": "", "post_date": "2024-03-26T22:36:43.950000", "content": "

We also added interactive visualizations of the 3D models with Rerun.io
\nYou need to clone and run the notebook to see the visualization.

", "votes": 3, "replies": [] }, { "id": 2808692, "author_name": "Arthuryangjc", "author_url": "", "post_date": "2024-05-12T11:01:23.887000", "content": "

What a great competition!

", "votes": 1, "replies": [] }, { "id": 2744809, "author_name": "zacstewart", "author_url": "", "post_date": "2024-04-10T06:22:56.693000", "content": "

What is the relationship of the \"scene\" and \"dataset\" fields in the submission CSV and the directory structure in train/ and test/? The data page of the competition says \"[train/test]/*/*/images\" with two asterisks, presumably representing dataset and scene.

\n

However, in the dataset it seems that the directory structure only contains one level. Is that dataset or scene?

\n
$ tree -d \n\n├── test\n│   └── church\n│       └── images\n└── train\n    ├── church\n    │   ├── images\n    │   └── sfm\n    ├── dioscuri\n    │   ├── images\n    │   └── sfm\n    ├── lizard\n    │   ├── images\n    │   └── sfm\n    ├── multi-temporal-temple-baalshamin\n    │   ├── images\n    │   └── sfm\n    ├── pond\n    │   ├── images\n    │   └── sfm\n    ├── transp_obj_glass_cup\n    │   ├── images\n    │   └── sfm\n    └── transp_obj_glass_cylinder\n        ├── images\n        └── sfm\n
", "votes": 1, "replies": [ { "id": 2744970, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-10T08:54:56.617000", "content": "

This is a leftover from the previous competition directory structure. This year you are correct that dataset is the same as the scene.

", "votes": 4, "replies": [] } ] }, { "id": 2747521, "author_name": "Benjamin Atiemo", "author_url": "", "post_date": "2024-04-11T23:56:16.087000", "content": "

I really love competition like this, preparing to get in touch

", "votes": 0, "replies": [] }, { "id": 2728379, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-02T07:57:23.693000", "content": "

We have added Deep Image Matching toolbox example https://www.kaggle.com/code/lcmrll/dim-package-submission-example

", "votes": 2, "replies": [] }, { "id": 2834572, "author_name": "lozhang19", "author_url": "", "post_date": "2024-05-24T20:22:05.830000", "content": "

nice this is greate one

", "votes": 0, "replies": [] }, { "id": 2819531, "author_name": "Sennikov Andrey", "author_url": "", "post_date": "2024-05-17T15:10:18.537000", "content": "

What a great competition!

", "votes": 0, "replies": [] }, { "id": 2814018, "author_name": "Hao Chen Bupt", "author_url": "", "post_date": "2024-05-15T05:52:25.563000", "content": "

This competition is really interesting!

", "votes": 0, "replies": [] }, { "id": 2810451, "author_name": "hebailin", "author_url": "", "post_date": "2024-05-13T09:20:00.603000", "content": "

Can we skip using Colmap and instead use other 3D reconstruction frameworks？

", "votes": 0, "replies": [ { "id": 2810519, "author_name": "old-ufo", "author_url": "", "post_date": "2024-05-13T09:49:30.130000", "content": "

Of course you can.

", "votes": 0, "replies": [] } ] }, { "id": 2792014, "author_name": "huangxiao309", "author_url": "", "post_date": "2024-05-04T03:24:09.257000", "content": "

Why did I run the code without task issues, the commit file was generated normally, I also checked the commit file: if there is an error in the number of rows or columns, a null value, the data type of the value is incorrect, or the commit value is inconsistent with the expected value. But in the end, it shows a scoring error, how should this be solved, does anyone have a similar situation?

", "votes": 0, "replies": [] }, { "id": 2780303, "author_name": "600162-สิทธิโชค", "author_url": "", "post_date": "2024-04-28T06:12:55.950000", "content": "

I like this competition

", "votes": 0, "replies": [] }, { "id": 2759965, "author_name": "MaybeRichard", "author_url": "", "post_date": "2024-04-19T03:21:27.433000", "content": "

sounds good

", "votes": 0, "replies": [] }, { "id": 2753943, "author_name": "Hans Kumar", "author_url": "", "post_date": "2024-04-15T18:12:37.320000", "content": "

Are the images for a particular dataset all collected with with the same camera (e.g. are the intrinsics fixed for a dataset?)

", "votes": 0, "replies": [ { "id": 2754059, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-15T19:13:23.580000", "content": "

This is a dangerous assumption to make.

", "votes": 0, "replies": [ { "id": 2766277, "author_name": "Hans Kumar", "author_url": "", "post_date": "2024-04-21T15:39:25.903000", "content": "

It seems difficult / potentially ill posed to solve SFM when all images are captured from cameras with different intrinsics and we have no prior info on these parameters. Is the EXIF data generally an accurate enough initial value for SFM? I also would be surprised if there not at least some images that have exactly the same intrinsics. Does anyone have any more intuition on this?

", "votes": 1, "replies": [] } ] } ] }, { "id": 2727667, "author_name": "vlad7carvalho", "author_url": "", "post_date": "2024-04-01T21:59:58.797000", "content": "

Hello! I can't figure out if OpenCV library (open-source part) is allowed to be used in this competition. Does anyone know?

", "votes": 0, "replies": [ { "id": 2728378, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-02T07:52:56.047000", "content": "

Of course it is allowed. Literally all open source libraries are allowed (MIT, GPL, BSD3, Apache2, etc), unless they are non-commercial license.

", "votes": 1, "replies": [] } ] }, { "id": 2724666, "author_name": "zzx1999", "author_url": "", "post_date": "2024-03-31T03:17:32.437000", "content": "

sounds good

", "votes": 0, "replies": [ { "id": 2770571, "author_name": "Evan Zhou", "author_url": "", "post_date": "2024-04-23T22:44:32.663000", "content": "

sounds good

", "votes": 0, "replies": [] } ] }, { "id": 2719273, "author_name": "Amartya Pawar", "author_url": "", "post_date": "2024-03-27T15:38:45.477000", "content": "

Can anyone explain what is the rotation matrix and translation vector

", "votes": 0, "replies": [ { "id": 2719427, "author_name": "old-ufo", "author_url": "", "post_date": "2024-03-27T17:44:52.023000", "content": "

You can check here https://en.wikipedia.org/wiki/Camera_matrix

\n

Or here https://kornia.readthedocs.io/en/latest/geometry.camera.pinhole.html#pinhole-camera

", "votes": 0, "replies": [] } ] }, { "id": 2852495, "author_name": "", "author_url": "", "post_date": "2024-06-03T10:19:26.547000", "content": "", "votes": 0, "replies": [] }, { "id": 2852491, "author_name": "", "author_url": "", "post_date": "2024-06-03T10:15:58.910000", "content": "", "votes": 0, "replies": [] }, { "id": 2737811, "author_name": "", "author_url": "", "post_date": "2024-04-06T00:34:43.030000", "content": "", "votes": 2, "replies": [ { "id": 2813093, "author_name": "Ningqin Lu", "author_url": "", "post_date": "2024-05-14T15:15:51.083000", "content": "

+1 any answer?

", "votes": 0, "replies": [] } ] }, { "id": 2797606, "author_name": "", "author_url": "", "post_date": "2024-05-06T19:55:16.443000", "content": "

Thank you to COMPETITION HOST!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2715623": "Hi everyone,\n\nI would like to welcome you to the 2024 edition of the Image Matching Challenge! This is the third edition we held at Kaggle. Both of the previous editions were quite successful: you can check out the recap thread if you're curious. \n\nThis year's challenge we raise the difficulty level of the task even more. Same as a last year you'll be building 3D reconstructions from medium-sized image sets: up to 100 images. \n\nHowever, instead of dealing with a single nuisance factor (e.g. in plane rotation for the historical preservation data last year), you will be dealing with multiple of them in the same time. For example, the images are taken from different view points AND at different time of the day AND during a different season. Or, for example, the data is of nature origin (trees, plants, flowers), and contains repeated patterns.\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- Links to the previous versions of the challenge: [2023](https://www.kaggle.com/competitions/image-matching-challenge-2023), [2022](https://www.kaggle.com/competitions/image-matching-challenge-2022)\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- [Creating a simple submission on GPU](https://www.kaggle.com/code/oldufo/imc-2024-submission-example), using ALIKED local features and LightGlue matcher through Kaggle models, and DINOv2 for creating a shortlist. This notebook contains more advanced auxiliary functions and datasets which may help you interact with Colmap, the 3D reconstruction framework bundled with the competition.\n- **Updated** [Running the evaluation metric on training set](https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example)\n- **Updated** [Training 3D model visualization with Rerun.io](https://www.kaggle.com/code/oldufo/colmap-3d-visualization-with-rerun-io)\n- **Updated** [Deep Image Matching toolbox example](https://www.kaggle.com/code/lcmrll/dim-package-submission-example)\n\n**Important**: you can also look at the [winner notebooks at 2023](https://www.kaggle.com/competitions/image-matching-challenge-2023/leaderboard) competition. However, they will not work \"out-of-the-box\" because the pycolmap version changed between the competitions. \n\nThe main differences in interface are in:\n\n- Reconstruction option object is renamed: `mapper_options = pycolmap.IncrementalMapperOptions()` in old version vs current `mapper_options = pycolmap.IncrementalPipelineOptions()`;\n- The structure, which contains camera pose: old version uses `im.rotmat(), im.tvec` vs `im.cam_from_world.rotation.matrix(), im.cam_from_world.translation` in new version.\n\nWe plan to add more resources in the following days, so you may want to follow this thread.\n\n### Prize-eligibility\n\n- Any submission, using non-commercial licensed 3rd party code is not prize eligible. E.g. submissions using SuperGlue (or SuperPoint) are not prize-eligible. Use LightGlue instead, which is both open source, and provides better performance. \n- GPL licensed 3rd party code is OK. However, your own code should be Apache 2.0 licensed. \n\nBest of luck!\n\n~ The organizers", "2716833": "We have also added the metric code calculation here https://www.kaggle.com/code/fabiobellavia/imc2024-3d-metric-evaluation-example", "2718037": "We also added interactive visualizations of the [3D models with Rerun.io](https://www.kaggle.com/code/oldufo/colmap-3d-visualization-with-rerun-io)\nYou need to clone and run the notebook to see the visualization. ", "2808692": "What a great competition!", "2744809": "What is the relationship of the \"scene\" and \"dataset\" fields in the submission CSV and the directory structure in train/ and test/? The data page of the competition says \"[train/test]/\\*/\\*/images\" with two asterisks, presumably representing dataset and scene.\n\nHowever, in the dataset it seems that the directory structure only contains one level. Is that dataset or scene?\n\n```\n$ tree -d data\ndata\n├── test\n│ └── church\n│ └── images\n└── train\n ├── church\n │ ├── images\n │ └── sfm\n ├── dioscuri\n │ ├── images\n │ └── sfm\n ├── lizard\n │ ├── images\n │ └── sfm\n ├── multi-temporal-temple-baalshamin\n │ ├── images\n │ └── sfm\n ├── pond\n │ ├── images\n │ └── sfm\n ├── transp_obj_glass_cup\n │ ├── images\n │ └── sfm\n └── transp_obj_glass_cylinder\n ├── images\n └── sfm\n```", "2747521": "I really love competition like this, preparing to get in touch", "2728379": "We have added Deep Image Matching toolbox example https://www.kaggle.com/code/lcmrll/dim-package-submission-example", "2834572": "nice this is greate one", "2819531": "What a great competition!", "2814018": "This competition is really interesting!", "2810451": "Can we skip using Colmap and instead use other 3D reconstruction frameworks？", "2792014": "Why did I run the code without task issues, the commit file was generated normally, I also checked the commit file: if there is an error in the number of rows or columns, a null value, the data type of the value is incorrect, or the commit value is inconsistent with the expected value. But in the end, it shows a scoring error, how should this be solved, does anyone have a similar situation?", "2780303": "I like this competition", "2759965": "sounds good", "2753943": "Are the images for a particular dataset all collected with with the same camera (e.g. are the intrinsics fixed for a dataset?)", "2727667": "Hello! I can't figure out if OpenCV library (open-source part) is allowed to be used in this competition. Does anyone know?", "2724666": "sounds good", "2719273": "Can anyone explain what is the rotation matrix and translation vector ", "2852495": "", "2852491": "", "2737811": "", "2797606": "Thank you to COMPETITION HOST!" } }