About Multimodal Semantic Segmentation:

With the rise of multimodal remote sensing (RS) and cutting-edge AI technologies, high-resolution ISA mapping is entering a new era. ISASeg, specifically designed for large-scale multimodal super-resolution segmentation, integrates diverse data sources to provide complementary and comprehensive information across modalities and spatial resolutions. ISASeg pairs globally available 10m Sentinel-1/2 imagery with 1m super-resolved images and finely annotated ISA labels, totaling over 4.26 billion labeled pixels across diverse urban environments. Designed to fuel AI-driven high-resolution ISA mapping, ISASeg enables fine-scale, long-term land cover analysis without the need for costly Very High Resolution (VHR) data. By supporting the development of scalable, super-resolution segmentation models, this dataset bridges the gap between medium-resolution satellites and fine urban-scale precision, offering a powerful foundation for future research and real-world applications in urban planning and environmental monitoring.

This contest is organized in conjunction with the 15th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS).

All participants have to submit:

1. Their segmentation results via the Kaggle platform before the specified deadline. The evaluation process is divided into two phases: Validation and Testing. During the Validation phase, both labeled training samples and unlabeled validation samples will be provided. Participants are required to submit their results through the following link: https://www.kaggle.com/competitions/whisper-2025-isa-seg-phase-1. The Testing phase will be announced and released subsequently. Final rankings will be determined based on the combined performance in both phases.
2. Each submission must also include the following information: 1) Group ID; 2) Affiliation; 3) List of team members (maximum of three, including the corresponding member, and emails for all members).
3. The report descripting their developed methods and models to the WHISPERS paper review system. A special session will be organized for accepted papers, which will also be included in the WHISPERS proceedings.

In addition, winning teams will receive an official award certificate and will be invited to co-author an IEEE JSTARS paper summarizing the outcomes of this challenge.

Note: Each team is limited to a maximum of three members. Any submission listing more than three participants will be deemed invalid. Furthermore, winning teams are required to provide the complete runnable code used to re-generate their final results. The codes will be freely and openly available with the subsequent IEEE JSTARS paper. The results produced by the submitted code must match those previously submitted; otherwise, the entry will be disqualified, and the award certificates will be revoked.

Reference: Jie Deng, Danfeng Hong, Chenyu Li, Naoto Yokoya. “Joint Super-Resolution and Segmentation for 1-m Impervious Surface Area Mapping in China’s Yangtze River Economic Belt.” arXiv preprint arXiv: 2505.05367 (2025).