**Updated!**
**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**.
**September 27, 2024**: We are happy to announce the winners of the challenge following the verification of the source code and technical report:
  i. Rank 1 ($1000): **Team DoubleY**
 ii. Rank 2 ($500): **Team BDASL**
iii. Rank 3 ($200): **Team Big Ben**
Congratulations to the winners!

**September 03, 2024**: The details of the final submission of the technique report and source code are released in discussion!
In 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.


## Introduction 
Building 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.

Currently, 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.

This 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.