Benchmarking and Enhancing VLM for Compressed Image Understanding
Zifu Zhang, Tongda Xu†, Siqi Li, Shengxi Li†, Yue Zhang, Mai Xu, Yan Wang†
Paper Code

We present a comprehensive benchmark designed to evaluate the ability of VLMs to understand and process compressed images. Based on this, we propose a lightweight VLM adaptor that enhances VLM performance on compressed images across diverse codecs and bitrate levels.

Machines Serve Human: A Novel Variable Human-machine Collaborative Compression Framework
Zifu Zhang, Shengxi Li†, Xiancheng Sun, Mai Xu, Zhengyuan Liu, Jingyuan Xia
Paper Code

We propose the first successful attempt by a novel collaborative compression method based on the machine-vision-oriented compression, named as diffusion-prior based feature compression for human and machine visions (Diff-FCHM).

Hierarchical Semantic Compression for Consistent Image Semantic Restoration
Shengxi Li (Supervisor), Zifu Zhang, Mai Xu, Lai Jiang, Yufan Liu, Ce Zhu
Paper Code

We propose a novel hierarchical semantic compression (HSC) framework that purely operates within intrinsic semantic spaces from generative models, which is able to achieve efficient compression for consistent semantic restoration.

FC-FORMER: Efficient Feature Coding for Machines via a Hybrid CNN-Transformer Architecture
Zhengyuan Liu*, Zifu Zhang*, Shengxi Li†, Tao Xu, Mai Xu, Xin Deng
Paper

We propose in this paper a novel feature compression architecture based on a hybrid of Transformer and convolutional neural network (CNN) architectures, thus named as FC-Former that enhances the capability of feature extraction, achieving the state-of-the-art performances.

Continuous Patch Stitching for Block-wise Image Compression
Zifu Zhang, Shengxi Li†, Henan Liu, Mai Xu, Ce Zhu
Paper Code

We propose a novel continuous patch stitching (CPS) framework for block-wise image compression that is able to achieve seamlessly patch stitching and mathematically eliminate block artefact, thus capable of significantly reducing the required computing resources when compressing images.

Hybrid Single Input and Multiple Output Method For Compressing Features Towards Machine Vision Tasks
Zifu Zhang, Shengxi Li, Tie Liu, Mai Xu, Tao Xu, Zhenyu Guan
Paper Code

This paper introduces a simple yet effective architecture called hybrid single input and multiple output (H-SIMO) for VCM, which can significantly reduce the redundancy across scales of features.