Solar Filament Segmentation Challenge 2026

Introduction

The "Solar Filament Segmentation Challenge 2026", organized by the Earth-Space Artificial Intelligence Research (ESAIR) Laboratory, focuses on developing computer vision algorithms to automatically segment solar filaments from GONG H-Alpha observation images. The goal is to generate accurate segmentation masks for each solar filament, including the smallest structures, while minimizing the inclusion of non-filament regions.

The competition emphasizes the precision and reliability of filament morphological delineation across various solar observation datasets. The primary evaluation metric is the Panoptic Quality (PQ) index, which combines segmentation accuracy with penalties for over-fragmentation and over-merging, as well as the computational efficiency of the proposed method.

Target Audience

• Undergraduate and graduate students.
• Researchers and professionals in the fields of computer vision, machine learning, and space science.

Benefits

• Opportunity to apply advanced computer vision and machine learning algorithms.
• Contribution to space science research and space weather forecasting.
• Winning teams will be announced at the IEEE BigData Conference.

Requirements

• Develop a solar filament segmentation algorithm.
• Any computational method can be used, including traditional image processing techniques, machine learning methods, and deep neural network architectures.
• Ensure the accuracy and efficiency of the algorithm.

How to Register

• Access the Google Form to complete the registration.
• Prepare a technical report (4 pages) and a public source code repository (GitHub, GitLab, Bitbucket).

Important Timeline

• July 10, 2026: Competition begins.
• November 15, 2026: Deadline for submitting technical reports and solutions.
• November 30, 2026: Announcement of winning teams.
• December 14-17, 2026: (Optional) IEEE BigData Conference in Phoenix, AZ, where winners will be announced.