Journal · 2025
ICNN Based Distributed Optimization of 3D Point Cloud Quality for Real-Time Physical Space Sharing
Abstract
Hybrid style metaverses, integrating physical and virtual spaces, face a critical challenge in managing shared 3D object quality across multiple users with diverse preferences and limited network resources. This paper addresses the problem of allocating limited bandwidth for transmitting point cloud representations while maximizing overall user satisfaction. We propose a distributed optimization method that dynamically adjusts 3D object quality based on contextual importance, available resources, and user preferences. Our approach uses Input Convex Neural Networks (ICNN) to model user utility functions and employs the Alternating Direction Method of Multipliers (ADMM) for distributed optimization. Key advantages include scalability, adaptability, and improved quality of experience. Evaluation using real-world data captured by our team and open datasets demonstrate significant improvements in user satisfaction and resource utilization compared to baseline approaches. Our method achieves 93-94.6% accuracy in modeling user utility and shows up to 60% faster convergence for scenarios with 30 users, contributing to the balance between high-fidelity representation and efficient data management in hybrid-metaverses.
Research Note
In a hybrid metaverse that merges physical and virtual spaces, managing the quality of the shared 3D objects is where the implementation becomes hard. Users differ in what they consider important and network resources are finite, so delivering maximum quality to everyone is not an option.
The problem is formulated here as allocating limited bandwidth for transmitting point cloud representations so as to maximise overall satisfaction. Object quality is adjusted dynamically according to three factors, the contextual importance of the object at that moment, the resources currently available, and the preferences of the user.
What makes the allocation difficult to design is that user satisfaction cannot simply be written down as a formula. Input Convex Neural Networks are used to learn the utility function from data instead. Because an ICNN is constrained so that its output is convex in its input, it satisfies two requirements at once, learning the shape of utility from measurements while keeping the subsequent resource allocation a tractable optimization problem. Distributed optimization is carried out with ADMM, which decomposes the problem, solves the parts and reconciles them into a global solution.
Evaluation on both data captured by the team and open datasets gave 93 to 94.6% accuracy in modelling user utility and up to 60% faster convergence in scenarios with 30 users, improving on baseline approaches in both satisfaction and resource utilisation.
Related Projects
