With the advancement of intelligent driving and computer vision systems, high-quality and diverse rainy-day images are crucial for training robust visual systems. However, traditional image generation models rely on large-scale centralized datasets, which raises serious issues such as data privacy, security, and data silos when handling geographically distributed data. To address these challenges, this paper proposes a novel federated learning-based framework for collaboratively training a global flow matching model capable of generating high-quality rainy-day images with regional characteristics while preserving privacy. The framework utilizes the FedAvg algorithm to distribute the flow matching model to local servers for regional training, capturing unique geographical features. After training, the model parameters are securely aggregated to a central server under encryption protection, forming a powerful global model through weighted averaging. Through rigorous comparisons with the TPSeNCE framework, experimental results demonstrate that our method successfully achieves distributed collaborative model training without sacrificing image quality or model performance, offering a new approach for scalable generative modeling in distributed environments.

Owing to its rapid progress and broad application prospects, few-shot action recognition has attracted considerable interest. However, current methods are predominantly based on limited single-modal data, which does not fully exploit the potential of multimodal information. This paper presents a novel framework that actively identifies reliable modalities for each sample using task-specific contextual cues, thus significantly improving recognition performance. Our framework integrates an Active Sample Inference (ASI) module, which utilizes active inference to predict reliable modalities based on posterior distributions and subsequently organizes them accordingly. Unlike reinforcement learning, active inference replaces rewards with evidence-based preferences, making more stable predictions. Additionally, we introduce an active mutual distillation module that enhances the representation learning of less reliable modalities by transferring knowledge from more reliable ones. Adaptive multimodal inference is employed during the meta-test to assign higher weights to reliable modalities. Extensive experiments across multiple benchmarks demonstrate that our method significantly outperforms existing approaches.