The Hidden Perception Constraint in Task-Aware Compression

2026-09-28

AI Summary

This study addresses how implicit perceptual constraints in task-aware compression affect the trade-off between classification utility and reconstruction quality. By integrating neural compression, information theory, and statistical testing theory, this work identifies and quantifies latent perceptual constraints within joint reconstruction-classification tasks. To resolve decision boundary misalignment, it proposes an optimization strategy based on target distribution matching that maximizes classification utility while minimizing bitrate. The contributions reveal naturally occurring perceptual constraint mechanisms inherent to task-aware compression. Furthermore, the proposed distribution matching scheme effectively calibrates classifier decision boundaries, substantially improving classification accuracy and achieving synergistic optimization of both classification performance and reconstruction quality under high compression ratios.

Abstract

With the recent advancements of neural compressors, explicitly incorporating perception constraints into the design of compression schemes has gained significant attention. Traditionally, these perception constraints ensure that the distribution of the reconstruction does not significantly deviate from the distribution of the source, thus attesting to the perceptual quality of the reconstruction. In this work, we uncover several perception constraints that are naturally present in task-aware compression. In particular, we consider a problem where the primary task is reconstruction and the secondary task is classification (i.e., a statistical test). We study this problem at varying levels of domain information available to us and discuss how to utilize the naturally emerging perception constraints to design rate-minimal compression schemes that also maximize the utility of our secondary task. We show that in this setting, if the decision boundaries of the classifier are ill-defined (mismatch) for our source distribution, then matching onto a target distribution enhances our classification accuracy.

Problem

• task-aware compression
• perception constraint
• rate-distortion
• classification
• distribution mismatch

Innovation

• Task-Aware Compression
• Perception Constraint
• Rate-Minimal Compression
• Distribution Matching
• Classification Utility