2026 GeoAI Arctic Challenge

Competition Rules

Eligibility

The competition is open to students, academics, industry professionals, and independent participants.

Participation is organized at the team level:

• Each team may include 1 to 5 members.
• Each team must designate one Team Leader.
• The Team Leader manages registration and is the main contact for the organizers.
• Each team may register only once.
• Each participant must have a Hugging Face account.
• Each team member must provide their exact Hugging Face username during registration.
• Multiple registrations representing the same team are not permitted.

Team membership should remain unchanged after registration unless explicitly approved by the organizers.

We strongly encourage and will give additional credit to teams composed of individuals with diverse backgrounds (e.g., Earth science + AI).

To participate in the challenge, teams must first register and be approved through the official Hugging Face competition page.

Registration

1. Form your team and select a Team Leader.
2. Create a Hugging Face account for each team member.
3. Complete the team registration form here.
4. Provide the team name, Team Leader contact information, team member names and affiliations, and the exact Hugging Face username for every team member.
5. The organizers will review registrations and verify eligibility.

After registration:

1. Visit the official competition page.
2. Log in with your Hugging Face account.
3. Wait for organizer approval.
4. Approved Hugging Face usernames will be added to the competition registration list.
5. Once approved, registered team members can submit through New submission and view shared team history through My submissions.

Access requests are reviewed manually. Users who are not listed as approved participants can browse the competition pages and leaderboard, but cannot upload submissions.

Team Accounts and Shared Submissions

The competition uses Hugging Face usernames to identify registered participants.

Registered users mapped to the same team share:

• My submissions
• New submission access
• The daily submission limit
• One leaderboard entry

All approved team members can submit for the team. Team names are managed by the organizers based on the registration form.

Submission Rules

• Submissions must be generated by automated model outputs.
• Manual labeling of the hidden test set is not permitted.
• Predictions must follow the required COCO results JSON format.
• Each prediction must use category_id: 1.
• Each mask must be encoded as compressed COCO RLE.
• Each score must be numeric and in [0, 1].
• Empty submissions are valid and mean no predicted RTS instances.
• Images with no predictions are valid.
• Multiple predictions per image are valid.
• The official metric uses the top-scoring detections per image according to maxDets=10.
• You may submit more than 10 predictions for an image, but predictions beyond the top 10 by score do not improve the official metric.

The competition platform will enforce settings such as maximum submissions per day, maximum upload size, leaderboard behavior, final ranking metric, and displayed leaderboard columns.

Data Usage

The dataset is provided for research and competition purposes within this challenge.

Redistribution of the dataset or derivatives outside the challenge requires permission from the original data owners and curators.

Participants may use external data, but external data usage must be documented in the method description or technical report.

Do not share labels, private data, submissions, or predictions across teams.

Attribution

Works or publications using this dataset should cite:

• Yang et al. (2023), the source RTS dataset paper.
• Li et al. (2025), the related multimodal GeoAI RTS modeling paper.

Full citation information is available on the Resources page.

Code and Report Requirements

All teams must submit a valid COCO results-format submission.json through the competition platform.

Top-performing teams must submit code for verification to be eligible for awards. They may also be invited to contribute to an outcome publication and may be asked to provide:

• A short technical report of 2 to 4 pages.
• Final inference code or a container to verify results.
• Instructions for reproducing predictions on the test set.
• Training scripts, if available.

The technical report should cover:

• Team members, affiliations, and country.
• Model architecture.
• Training strategy, including losses, augmentations, optimizer, batch size, and hardware.
• Special techniques such as domain adaptation, self-training, ensembling, or band selection.
• Results and ablations.
• Insights, challenges, lessons learned, and possible future improvements.
• References to related work.

A report template will be provided to keep submissions consistent.

Winner Selection Criteria

Competition winners will be selected based on a combination of leaderboard performance and qualitative evaluation.

Leaderboard Performance (80%)

Leaderboard ranking will account for 80% of the final score. This component reflects the performance of each team’s final submission according to the official competition evaluation metric.

Qualitative Evaluation (20%)

The remaining 20% of the final score will be determined through a review conducted by the organizing committee. Evaluation will consider:

• Technical report quality, including clarity, completeness, and reproducibility;
• Novelty and innovation of the proposed methodology;
• Interdisciplinary collaboration, including the extent to which the team combines expertise from different disciplines (e.g., Earth science, remote sensing, computer vision, machine learning, or related fields).

The organizers may request additional materials, including inference code or containers, to verify reported results before determining the final rankings.

The final competition winners will be determined using the combined weighted score from the leaderboard performance and the qualitative evaluation.

Conduct and Disqualification

Teams are expected to follow the competition rules, Hugging Face platform terms, and any competition-specific conduct policies.

Teams may be disqualified at the organizers’ discretion for:

• Registering the same team multiple times.
• Circumventing participation or submission limits.
• Misusing the dataset.
• Sharing private labels, predictions, or submissions across teams.
• Attempting to game the leaderboard.
• Submitting manually labeled hidden test predictions.

Important Dates

Event	Date
Competition opens	July 1, 2026
Dataset released	July 1, 2026
Benchmark phase starts	August 15, 2026
Final submission deadline	January 31, 2027
Winners announced	March 1, 2027