Paper accepted at WACV 2025

Our work “When Cars meet Drones: Hyperbolic Federated Learning for Source-Free Domain Adaptation in Adverse Weather”, has been accepted at WACV 2025 (Round 1 acceptance, 12.1% submissions), which will be held in Tucson, Arizona, from February 28 to March 4, 2025.

TL;DR: We address the unique challenges of Federated Learning (FL) for autonomous vehicles, in scenarios involving adverse weather conditions and the coexistence of different autonomous agents like cars and drones 🚗🚁. We introduce a novel federated semantic segmentation approach using a batch-norm weather-aware strategy to dynamically adapts to different weather conditions, hyperbolic prototypes to ensure consistent training across car and drone clients, and a queue aggregation strategy to stabilize the federated adaptation. To support our work, we also introduced 𝗙𝗟𝗬𝗔𝗪𝗔𝗥𝗘, the first semantic segmentation dataset specifically designed for aerial vehicles in adverse weather conditions.

Paper: https://arxiv.org/pdf/2403.13762

Code: https://github.com/LTTM/HyperFLAW