PORT: Perception-Oriented Image Compression with Real-Time Decoding

Wei Jiang, Bohao Feng, Wenqiang Wang, Bo huang, Lin Ding, Ronggang Wang

Abstract: This white paper presents PORT, a Perception-Oriented image compression framework with Real-Time decoding. PORT is our approach for the image compression track at CLIC 2025. To enhance perceptual quality, we incorporate both semantic and patch-wise adversarial losses to generate realistic textures, and employ a region-of-interest (ROI) mask to guide bit allocation across different regions. To accelerate decoding, PORT builds upon the DCVC-RT architecture, while introducing more advanced entropy models to capture long-range correlations. Our team is PKUSZ-AliMerlin.

Team Name: PKUSZ-AliMerlin

Submission Number: 1