{
  "fetched_at": "2026-10-19T07:26:10Z",
  "source_url": "https://www.kaggle.com/competitions/building-extraction-generalization-2024/data",
  "pages": [
    {
      "name": "data-description",
      "content": "We present a comprehensive collection of datasets designed to encompass a broad spectrum of land use classifications in Japan. This collection includes a diverse selection of satellite imagery and detailed mappings of building footprints across a range of locales, including central, urban, suburban, and rural areas. The building footprint information provides precise local coordinates for the vertices of the building outline polygons, formatted in COCO standard (https://cocodataset.org/#home). \nThe entire dataset is systematically organized into two parts: training set and test set. Figure 1 delineates the geographical coverage of the training and test sets, highlighting the concentration of the training set in the Tokyo vicinity, while the test sets are strategically dispersed across Japan. This distribution is designed to rigorously evaluate the model's generalization abilities.  Learning strategies such as transfer learning and domain adaptation are encouraged to use.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4876373%2F549622dd3280963b81ff4ed992521a6f%2F2.png?generation=1713961058316432&alt=media)\n## Training dataset\nThe training set data uses 0.3-meter Google earth satellite images complemented by meticulously manually annotated building outlines. The training set data was collected from:\n- **Susono**: Represents rural areas and contains 1,113 images, 18,159 building footprints\n- **Hachioji**: Represents suburban areas and contains 1,234 images, 30,841 building footprints\n- **Setagaya**: Represents urban areas and contains 1,272 images, 57,132 building footprints\n- **Hachioji**: Represents downtown areas and contains 1,098 images, 63,979 building footprints\n\nWith a ratio of 8:2 the training data was divided into a training set and a validation set\n## Test dataset\nThe imagery and building annotations for both test sets derive from the open-source Japanese 3D city model, the Plateau project (https://www.mlit.go.jp/plateau/) enhanced with manual adjustments following visual inspection. All test images were randomly selected from 42 cities in Japan, but a balance of different types of areas was maintained. A total of 250 images were taken from each region, totaling 1,000 images.\n\n\nWe assure that privacy concerns have been fully addressed in the compilation of this data. \n\n## Folders and files\n- **train/** - a folder where the training data is stored. Where the image folder holds 3,784 training images and the train.json file records the annotations in coco format.\n- **val/** - a folder where the validation data is stored. Where the image folder holds 933 validation images and the train.json file records the annotations in coco format.\n- **test/** - the 1,000 images of the challenge test set.\n- **sample_submission.csv** - a (truncated) sample submission file in the correct format. It includes 14 detections over (only) 6 images.\n\n## Submission Format \nThe extraction results of all images should to be saved in one single CSV file, with each line representing the results of one image. Each line must include the image id (***ImageID***) and a list of coordinates for the extraction results (***Coordinates***). The *ImageID* should match the image name but does not need to be four digits long (e.g., an image named 0001.jpg would have an *ImageID* of 1). \n\nThe *Coordinates* column should contain a list of coordinates for the extracted mask for each building. These coordinates should be formatted in [(x0, y0), (x1, y1), (x2, y2), ... , (xn, yn)]. To facilitate reading the coordinate column with *Pandas* , the coordinates needs to be stored as a string (This means the square brackets [ ] should be enclosed in quotation marks \" \"). For detailed examples, refer to the ***sample_submission.csv***. \n\nIf no building is extracted from a particular image, record an empty list (\"[ ]\" or []) in the *Coordinate* column instead of skipping the row. Therefore, your result file should contain 1000 lines, corresponding to the number of images in the test set.\n\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "IEEE BigData Cup 2024",
      "content": "This competition is part of the IEEE BigData Cup Challenge at the IEEE International Conference on BigData. A panel will review the top submissions for the competition based on the description of the methods provided. From this, a subset may be invited to present their results at the workshop of the conference. Attending the workshop is not required to participate in the competition; however, only teams that are attending the workshop will be considered to present their work.\n\nYou can see a list of all of the IEEE BigData 2024 competitions [here](https://www3.cs.stonybrook.edu/~ieeebigdata2024/).\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "**One account per participant**\nYou cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.\n\n**No private sharing outside teams**\nPrivately sharing code or data outside of teams is not permitted.\n\n**Team mergers**\nTeam mergers are allowed and can be performed by the team leader. In order to merge, the combined team must have a total submission count less than or equal to the maximum allowed as of the Team Merger Deadline. The maximum allowed is the number of submissions per day multiplied by the number of days the competition has been running.\n\n**Team limits**\nThe maximum team size is 5.\n\n**Submission limits**\nYou may submit a maximum of 5 entries per day.\nYou may select up to 2 final submissions for judging.\n\n**Competition timeline**\nCompetition Timeline dates (including Entry Deadline, Final Submission Deadline, Start Date, and Team Merger Deadline, as applicable) are reflected on the competition’s Overview > Timeline page.\n\n**Data access and use**\nCompetition Use, Non-Commercial and Academic Research Only.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "**Updated!**\n**September 27, 2024**: The top five teams are invited to submit papers. Papers must follow the IEEE conference template format as described here: [Manuscript Templates for Conference Proceedings](https://www.ieee.org/conferences/publishing/templates.html). The deadline for the paper submission is **October 15**.\n**September 27, 2024**: We are happy to announce the winners of the challenge following the verification of the source code and technical report:\n  i. Rank 1 ($1000): **Team DoubleY**\n ii. Rank 2 ($500): **Team BDASL**\niii. Rank 3 ($200): **Team Big Ben**\nCongratulations to the winners!\n\n**September 03, 2024**: The details of the final submission of the technique report and source code are released in discussion!\nIn addition, due to the contact limitations of Kaggle, please fill out the [**personal information form**](https://forms.gle/9myaaQxUWaskfZxS7) as soon as possible to facilitate follow-up contact.\n\n\n## Introduction \nBuilding footprint data, essential for identifying the exact location and shape of buildings, has become a foundation of urban Geographic Information Systems (GIS). With the advent of advanced artificial intelligence and remote sensing technologies, the automatic extraction of building footprints from high-resolution satellite imagery has garnered considerable interest. Techniques like Convolutional Neural Networks (CNN) and Transformers have shown remarkable proficiency in delineating building footprints from complex satellite images swiftly and accurately. Nonetheless, the inherent diversity of buildings in terms of shape, color, and size, alongside regional variations in environmental and land-use characteristics, presents substantial challenges. Extraction algorithms must be both flexible and adept, capable of discerning and recognizing building features across diverse settings. This necessitates algorithms that not only exhibit strong generalization capabilities but also maintain high accuracy levels, particularly for nationwide extraction tasks.\n\nCurrently, accessing a broad and diverse set of training data faces significant challenges, both financially and logistically, especially within real-world contexts. Ground-truth data, essential for model training and validation, are both rare and invaluable, often confined to specific locales. Addressing these constraints requires innovative methodologies that either work with limited training samples or effectively leverage available open-source data to achieve precise national-scale building footprint extraction and ensure consistent model performance across different areas.\n\nThis competition embarks on this challenge by utilizing a building footprint dataset from the Tokyo area as the primary training set, with plans to extend testing to other Japanese regions. This approach aims to inspire the development of models with robust generalization capabilities, capable of overcoming the hurdles of automatic building footprint detection and extraction across various landscapes. We advocate for the exploration of cross-domain generalization techniques, like fine-tuning and unsupervised learning, to enhance model performance and adaptability. Additionally, we encourage the use of public data sources, such as OpenStreetMap (OSM), to enrich training datasets. Overcoming this challenge signifies the creation of a novel approach for efficient, cost-effective, and precise building footprint extraction at a national level with minimal regional data, showcasing its potential applicability worldwide.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "## Evaluation metrics\n\nThe evaluation of results utilizes the object-wise F1-score. As depicted in Figure 1, the criteria for a correctly extracted building are presented, where the ground truth (GT) is represented by the blue polygon, and the extracted footprint by the yellow polygon. The True Positive (TP) is determined through the Intersection over Union (IoU) of these two polygons, specifically, the ratio of their overlapping area to their combined area, as illustrated in Figure 1a. A prediction is deemed accurate if the IoU between the two polygons is 0.5 or higher, as demonstrated in Figure 1c. Conversely, an IoU below 0.5, or the absence of a corresponding GT, is classified as a wrong extraction (False Negative, FN). Additionally, any overlooked extracted building is categorized as a False Positive (FP). The F1 score can then be calculated based on the quantities of TP, FN, and FP, according to equation below. Ultimately, the team that achieves the highest F1 score, based on the average evaluation result of test set without weight, will be declared the winner of the competition.\n\nPrecision = TP / (TP + FP)\nRecall = TP / (TP + FN)\nF1 = 2 * Recall * Precision / (Recall + Precision)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4876373%2Fe48685a5e72ee69a7dd45f0de47c4128%2F1.png?generation=1713959053396214&alt=media)\nFigure 1. Schematic representation of evaluation criteria for correctly extracted buildings.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Timeline",
      "content": "- **May 28, 2024**, Website of the challenge opens, and the dataset is revealed.\n- **June 1, 2024**: The submission portal opens. Participants can now register their team and submit their solutions for evaluation.\n- **August 31, 2024**: Deadline for submitting results.\n- **September ~~10~~ 15, 2024**: Deadline for submitting the source code & the detailed report of the solutions. End of the competition.\n- **September 25 - 30, 2024**: Announcement of winning teams and invitation for paper submission for the special track - during the IEEE BigData 2024 conference.\n- **October 15, 2024**: Deadline for the submission of the invited papers and final solutions.\n- **October 30, 2024**: Notification of paper acceptance.\n- **November 15, 2024**: Camera-ready version of accepted papers. \n- **December 15 –  18, 2024**: IEEE BigData 2024 Conference, Washington, US.\n\nAll deadlines are at 11:59 PM UTC on a corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Supplementary",
      "content": "## Task detail \n\nParticipants are tasked with crafting an algorithm that boasts outstanding generalization capabilities, utilizing training data confined to a particular region in Japan. This algorithm must possess the precision and autonomy to identify and delineate building footprints from satellite imagery, which geographically diverges from the training dataset. The core of this extraction endeavor hinges on pinpointing the exact coordinates of each building's boundary vertices with high accuracy. \n\nWe place no limitations on the algorithmic approaches’ participants may choose to adopt. We strongly advocate for the exploration of a broad spectrum of innovative algorithms, cutting-edge review-stage techniques, or established methods across supervised, weakly supervised, or unsupervised learning paradigms. Additionally, participants are encouraged to venture into the amalgamation of open-source data with established public datasets, such as OpenStreetMap (OSM), within their research and development activities. It's important to note, however, that the creation and utilization of any manually generated supplementary datasets are strictly forbidden.\n\n## Further reading\n- Chen, S., Ogawa, Y., Zhao, C., & Sekimoto, Y. (2023). Large-scale individual building extraction from open-source satellite imagery via super-resolution-based instance segmentation approach. ISPRS Journal of Photogrammetry and Remote Sensing, 195, 129-152.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Organizers",
      "content": "- Organizer 1\n**Shenglong Chen**\nProject researcher, Ph.D., Center for Spatial Information Science, The University of Tokyo\nContact information: chen-sl@csis.u-tokyo.ac.jp\n\n\n- Organizer 2\n**Chenbo Zhao**\nDoctoral student, Department of Civil engineering, The University of Tokyo\nContact information: zhaocb@csis.u-tokyo.ac.jp\n\n- Organizer 3\n**Bhanu Prasad Chintakindi**\nDoctoral student, Department of Civil engineering, The University of Tokyo\nContact information: bhanu@csis.u-tokyo.ac.jp\n\n- Organizer 4\n**Yoshiki Ogawa**\nLecturer, Ph.D., Center for Spatial Information Science, The University of Tokyo\nContact information: ogawa@csis.u-tokyo.ac.jp\n\n- Organizer 5\n**Yoshihide Sekimoto**\nProfessor, Dr.Eng., Center for Spatial Information Science, The University of Tokyo\nContact information: sekimoto@csis.u-tokyo.ac.jp\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Join The IEEE Big Data Challenge",
      "content": "- After the competition phase is completed, an invitation for submitting the accompanying academic paper (4-10 pages) will be provided to the top 5 participants (may be reduced depending on participation) as ranked by the leaderboard. Accepted papers will be invited to give talks at the IEEE BigData 2024 on December 15-18, 2024 in Washington, USA.\n- Only teams that are attending the workshop will be considered to present their work and got the prize.\n- Three teams will be selected as the winners based on the results and submitted paper of the challenge and awarded the following prizes:\n    - 1st Place $ 1,000\n    - 2nd Place $ 500\n    - 3rd Place $ 200\n- The papers are expected to conform to the[ IEEE Computer Society Proceedings Manuscript Formatting Guideline](https://www.ieee.org/conferences/publishing/templates.html). Paper be submitted to the following link: https://wi-lab.com/cyberchair/2024/bigdata24/index.php \n- Peer reviewers will review the academic papers.\n- Important Dates\n    - September 25 - 30, 2024: Announcement of winning teams and invitation for paper submission for the special track during the IEEE BigData 2024 conference.\n    - October 15, 2024: Deadline for the submission of the papers.\n    - October 30, 2024: Notification of paper acceptance.\n    - November 15, 2024: Camera-ready version of accepted papers. \n    - December 15 – December 18, 2024: IEEE BigData 2024 Conference, Washington, USA.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Submission",
      "content": "## Contents in the technical report (Required):\n- Explanation of your method and tools\n- Evaluation of your method.\n- Detailed evaluation of your results based on factors such as inference speed, model size, etc.\n- Code and trained model links.\n- Error Analysis: Examples of failed attempts, efforts that did not go well.\n\n## Code Requirement\nSource code will also be required to be submitted, either through a publicly available repository on a Git-based version control hosting service such as GitHub for the final evaluation. All source codes are expected to be released as open-source software, utilizing some generally accepted licensing such as Apache License 2.0, GNU General Public License, MIT license, or others of similar acceptance by the Open Source Initiative.\n\n## Paper submission\n- After the competition phase is completed, a link for submitting the accompanying academic paper will be provided to the top 5 participants as ranked by the leaderboard described above.\n- Peer reviewers will review the academic papers.\n- The papers are expected to conform to the format set by the conference, which can be found at IEEE BigData CFP.\n\n## Contact\nIf you have questions, please feel free to contact the lead organizer chen-sl@csis.u-tokyo.ac.jp",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}