Progressive Learned Image Compression for Machine Perception

2025-12-23

AI Summary

Progressive image compression for machine perception must jointly optimize fine-grained scalability (FGS) and task-oriented rate-distortion performance. This paper introduces PICM-Net, the first learned codec supporting FGS via ternary-plane coding, enabling multi-quality-level decoding from a single bitstream. Our contributions are threefold: (1) the first integration of progressive compression with task-aware rate-distortion optimization; (2) an adaptive decoding controller conditioned on downstream-task confidence, enabling inference-time dynamic quality adjustment; and (3) a latent-space priority modeling framework with end-to-end joint training. Evaluated on ImageNet classification under bandwidth constraints, PICM-Net retains over 98.5% of the original model’s accuracy—substantially outperforming non-progressive baselines. This work establishes a new paradigm for efficient, adaptive machine-vision data transmission.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Learning & Optimization for CV
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Recent advances in learned image codecs have been extended from human perception toward machine perception. However, progressive image compression with fine granular scalability (FGS)-which enables decoding a single bitstream at multiple quality levels-remains unexplored for machine-oriented codecs. In this work, we propose a novel progressive learned image compression codec for machine perception, PICM-Net, based on trit-plane coding. By analyzing the difference between human- and machine-oriented rate-distortion priorities, we systematically examine the latent prioritization strategies in terms of machine-oriented codecs. To further enhance real-world adaptability, we design an adaptive decoding controller, which dynamically determines the necessary decoding level during inference time to maintain the desired confidence of downstream machine prediction. Extensive experiments demonstrate that our approach enables efficient and adaptive progressive transmission while maintaining high performance in the downstream classification task, establishing a new paradigm for machine-aware progressive image compression.

Problem

• Develops progressive image compression for machine perception tasks
• Introduces adaptive decoding to maintain downstream prediction confidence
• Establishes a new paradigm for machine-aware progressive compression

Innovation

• Progressive learned compression using trit-plane coding
• Adaptive decoding controller for dynamic inference adjustment
• Machine-aware prioritization for efficient progressive transmission