Hi all,
\nI'm new to computer vision field and I have an interest in reconstructing of 3D objects. In this competition, for the image matching part, it was clear to understand for me. After reading the explanation of their works from the top ranked teams, I've acknowledge the different approaches to match and cluster the images of the same object.
\nFor the reconstruction part, I've tried using Colmap GUI to reconstruct the images. However, I would like to implement in Python so that I can use Kaggle notebook/Google Colab for reconstruction. I've also tried PyColmap to implement but it doesn't have a proper documentation. Thus, it is very difficult for me to understand how it works.
\nTherefore, I would like to know how you guys reconstruct the 3D object from the matching images in this competition.
Thank you and congratulations to all the winners of the competition.
", "messages": [ { "id": "2308780", "postDate": "06/19/2023 08:00:31", "content": "Hi all,
\nI'm new to computer vision field and I have an interest in reconstructing of 3D objects. In this competition, for the image matching part, it was clear to understand for me. After reading the explanation of their works from the top ranked teams, I've acknowledge the different approaches to match and cluster the images of the same object.
\nFor the reconstruction part, I've tried using Colmap GUI to reconstruct the images. However, I would like to implement in Python so that I can use Kaggle notebook/Google Colab for reconstruction. I've also tried PyColmap to implement but it doesn't have a proper documentation. Thus, it is very difficult for me to understand how it works.
\nTherefore, I would like to know how you guys reconstruct the 3D object from the matching images in this competition.
Thank you and congratulations to all the winners of the competition.
", "rawMarkdown": "Hi all, \nI'm new to computer vision field and I have an interest in reconstructing of 3D objects. In this competition, for the image matching part, it was clear to understand for me. After reading the explanation of their works from the top ranked teams, I've acknowledge the different approaches to match and cluster the images of the same object. \nFor the reconstruction part, I've tried using Colmap GUI to reconstruct the images. However, I would like to implement in Python so that I can use Kaggle notebook/Google Colab for reconstruction. I've also tried PyColmap to implement but it doesn't have a proper documentation. Thus, it is very difficult for me to understand how it works.\nTherefore, I would like to know how you guys reconstruct the 3D object from the matching images in this competition. \n\nThank you and congratulations to all the winners of the competition.", "votes": null }, { "id": "2309793", "postDate": "06/19/2023 22:17:39", "content": "Here's an example notebook if you want to visualize the 3D reconstruction using pycolmap:
\nhttps://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis
Here's the official documentation of Colmap . Basically, Colmap stores all the matching points in its database (a local file) and try to jointly optimize for the 3D coordinates corresponding to those feature points and the camera pose parameters, in terms of minimizing the 2D reprojection error. This technique is called bundle adjustment.
\nIf you want to know how the math works, you can refer to the textbook Multiview Geometry, especifically part IV: N-View Geometry.
", "rawMarkdown": "Here's an example notebook if you want to visualize the 3D reconstruction using pycolmap: \nhttps://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis\n\nHere's the [official documentation of Colmap ](https://colmap.github.io/). Basically, Colmap stores all the matching points in its database (a local file) and try to jointly optimize for the 3D coordinates corresponding to those feature points and the camera pose parameters, in terms of minimizing the 2D reprojection error. This technique is called bundle adjustment. \n\nIf you want to know how the math works, you can refer to the textbook [Multiview Geometry](https://github.com/DeepRobot2020/books/blob/master/Multiple%20View%20Geometry%20in%20Computer%20Vision%20(Second%20Edition).pdf), especifically part IV: N-View Geometry.", "votes": null }, { "id": "2311291", "postDate": "06/21/2023 04:45:42", "content": "Thank you for sharing the amazing resources. It really helps me in self-studying the 3D reconstruction.
", "rawMarkdown": "Thank you for sharing the amazing resources. It really helps me in self-studying the 3D reconstruction.", "votes": null } ], "comments": [ { "id": 2309793, "author_name": "anonymousyuxiang", "author_url": "", "post_date": "06/19/2023 22:17:39", "content": "Here's an example notebook if you want to visualize the 3D reconstruction using pycolmap:
\nhttps://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis
Here's the official documentation of Colmap . Basically, Colmap stores all the matching points in its database (a local file) and try to jointly optimize for the 3D coordinates corresponding to those feature points and the camera pose parameters, in terms of minimizing the 2D reprojection error. This technique is called bundle adjustment.
\nIf you want to know how the math works, you can refer to the textbook Multiview Geometry, especifically part IV: N-View Geometry.
", "votes": null, "replies": [ { "id": 2311291, "author_name": "kaungmyatkyaw", "author_url": "", "post_date": "06/21/2023 04:45:42", "content": "Thank you for sharing the amazing resources. It really helps me in self-studying the 3D reconstruction.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2308780": "Hi all, \nI'm new to computer vision field and I have an interest in reconstructing of 3D objects. In this competition, for the image matching part, it was clear to understand for me. After reading the explanation of their works from the top ranked teams, I've acknowledge the different approaches to match and cluster the images of the same object. \nFor the reconstruction part, I've tried using Colmap GUI to reconstruct the images. However, I would like to implement in Python so that I can use Kaggle notebook/Google Colab for reconstruction. I've also tried PyColmap to implement but it doesn't have a proper documentation. Thus, it is very difficult for me to understand how it works.\nTherefore, I would like to know how you guys reconstruct the 3D object from the matching images in this competition. \n\nThank you and congratulations to all the winners of the competition.", "2309793": "Here's an example notebook if you want to visualize the 3D reconstruction using pycolmap: \nhttps://www.kaggle.com/code/leonidkulyk/eda-imc-3d-plots-interactive-vis\n\nHere's the [official documentation of Colmap ](https://colmap.github.io/). Basically, Colmap stores all the matching points in its database (a local file) and try to jointly optimize for the 3D coordinates corresponding to those feature points and the camera pose parameters, in terms of minimizing the 2D reprojection error. This technique is called bundle adjustment. \n\nIf you want to know how the math works, you can refer to the textbook [Multiview Geometry](https://github.com/DeepRobot2020/books/blob/master/Multiple%20View%20Geometry%20in%20Computer%20Vision%20(Second%20Edition).pdf), especifically part IV: N-View Geometry.", "2311291": "Thank you for sharing the amazing resources. It really helps me in self-studying the 3D reconstruction." }, "source": "meta" }