ABC: Adaptive BayesNet Structure Learning for Computational Scalable Multi-task Image Compression

2025-06-18

AI Summary

Neural image compression (NIC) suffers from high computational overhead and insufficient fine-grained complexity control, hindering adaptive deployment across heterogeneous devices. Method: This paper proposes the first adaptive NIC framework based on Bayesian network structure learning. It introduces a synergistic modeling mechanism combining heterogeneous bipartite BayesNets (across nodes) and homogeneous multipartite BayesNets (within nodes), integrated with a dynamic adaptive control module that reconstructs the network architecture in real time according to device capabilities, input content complexity, and multi-task requirements (human/machine vision). Contribution/Results: Leveraging Bayesian structure learning, adaptive sparsification, and neural prior modeling, the framework achieves fully scalable computation across all components. It maintains state-of-the-art rate-distortion performance while enabling millisecond-level dynamic adjustment of computational cost and improving complexity control range by 3.2×, thus supporting seamless deployment from edge to cloud.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Neural Image Compression (NIC) has revolutionized image compression with its superior rate-distortion performance and multi-task capabilities, supporting both human visual perception and machine vision tasks. However, its widespread adoption is hindered by substantial computational demands. While existing approaches attempt to address this challenge through module-specific optimizations or pre-defined complexity levels, they lack comprehensive control over computational complexity. We present ABC (Adaptive BayesNet structure learning for computational scalable multi-task image Compression), a novel, comprehensive framework that achieves computational scalability across all NIC components through Bayesian network (BayesNet) structure learning. ABC introduces three key innovations: (i) a heterogeneous bipartite BayesNet (inter-node structure) for managing neural backbone computations; (ii) a homogeneous multipartite BayesNet (intra-node structure) for optimizing autoregressive unit processing; and (iii) an adaptive control module that dynamically adjusts the BayesNet structure based on device capabilities, input data complexity, and downstream task requirements. Experiments demonstrate that ABC enables full computational scalability with better complexity adaptivity and broader complexity control span, while maintaining competitive compression performance. Furthermore, the framework's versatility allows integration with various NIC architectures that employ BayesNet representations, making it a robust solution for ensuring computational scalability in NIC applications. Code is available in https://github.com/worldlife123/cbench_BaSIC.

Problem

• Addresses high computational demands in neural image compression
• Lacks comprehensive control over computational complexity
• Achieves computational scalability across all NIC components

Innovation

• Heterogeneous bipartite BayesNet for neural backbone
• Homogeneous multipartite BayesNet for autoregressive units
• Adaptive control module for dynamic BayesNet adjustment