BRIGHT Challenge: Advancing All-Weather Building Damage Mapping to Instance-Level
Overview
Natural and man-made disasters cause severe damage to urban infrastructure worldwide. Rapid and accurate mapping of building damage is essential for effective disaster response, resource allocation, and recovery planning. While semantic-level (pixel-wise) damage mapping has been extensively studied, instance-level damage mapping — which identifies and assesses each individual building separately — remains a critical yet underexplored challenge.
The BRIGHT Challenge invites participants to develop methods for instance-level building damage mapping using the BRIGHT dataset, a globally distributed multimodal benchmark featuring paired very-high-resolution (VHR) optical and synthetic aperture radar (SAR) imagery collected after real-world disaster events. The goal is to simultaneously detect, delineate, and classify each damaged building at the instance level.
This challenge is part of Monitoring the World Through an Imperfect Lens (MONTI) in conjunction with CVPR 2026 Conference and aims at advancing geoscience and remote sensing research through open benchmarks and competitive evaluation.
Task Description
Given a pair of pre-disaster optical and post-disaster SAR images, participants must produce an instance-level building damage map that:
- Detects and delineates each individual building footprint as a distinct instance.
- Assigns each detected building a damage level according to the standard classification: Intact, Damage, and Destroyed.
- Works under the all-weather constraint: methods should be capable of handling both optical and SAR inputs or their multimodal combination, addressing cases where optical imagery is unavailable due to cloud cover or night-time conditions.
Dataset
This challenge is based on the BRIGHT dataset, published in Earth System Science Data (ESSD), 2025.
The BRIGHT dataset covers multiple disaster types (earthquakes, floods, hurricanes, wildfires, etc.) across diverse geographic regions. Each scene provides:
- Pre-event optical imagery (VHR, sub-meter resolution)
- post-event SAR imagery (co-registered with optical data)
- Instance-level building annotations with per-instance damage level labels
Data Splits
| Split | Scenes | Building Instances | Labels Provided |
|---|---|---|---|
| Training | [N_TRAIN] | [N_INSTANCES_TRAIN] | Yes |
| Validation | [N_VAL] | [N_INSTANCES_VAL] | Withheld |
| Test | [N_TEST] | [N_INSTANCES_TEST] | Withheld |
Important Dates
All deadlines are at 23:59 AoE (Anywhere on Earth, UTC−12).
-
[DATE]
Challenge Launch & Data Release
Training and validation data become available for download. -
[DATE]
Development Phase Opens
Submission system opens; participants may submit results on the validation set. -
[DATE]
Test Data Release
Test images (without labels) are released for final evaluation. -
[DATE]
Final Submission Deadline
Deadline for final predictions on the test set. -
[DATE]
Code & Technical Report Submission
Top-ranked teams must submit their code and a short technical report. -
[DATE]
Results Announcement
Final leaderboard and winner notification. -
[DATE]
Workshop / Awards Ceremony
[CONFERENCE / WORKSHOP PLACEHOLDER]
Rules
- Participation is open to individuals and teams worldwide. No registration fee is required.
- Teams may use the BRIGHT training set and any publicly available pre-trained models or auxiliary data. Any additional data used must be declared in the technical report.
- The test labels are never released during the competition period.
- Each team may submit at most 10 submissions per day during the development phase.
- Each team can make a total of 10 submissions in the test phase.
- Teams finishing in the top ranks must provide reproducible code. Failure to do so will result in disqualification.
- Organizers reserve the right to disqualify submissions that violate these rules or show signs of test-label leakage.
Result Submission
Predictions should be submitted as COCO-format JSON following the format specification in the starter kit. Detailed submission instructions and example scripts are provided in the baseline repository.
Awards & Prizes
Top-ranked teams will be invited to present their solutions at MONTI of CVPR 2026.
Organizers
For inquiries, please contact: Qschrx@gmail.com
Citation
If you use the BRIGHT dataset or this challenge in your research, please kindly cite: