Multimedia Systems Computer Networks and Communications

Dual feature modulation for hybrid neural representation-based video compression

Xu, Weitao, Fang, Yi, Liu, Yuan, Ling, Qiang
Springer Nature · 2026-11-14 · DOI 10.1007/s00530-026-02792-y

Abstract

Neural representation-based video compression simplifies the coding pipeline and enables fast decoding by mapping video contents into neural network parameters. However, existing methods often suffer from performance degradation in the reconstruction of high-frequency details. In this paper, we propose Dual-Modulated Hybrid Neural Representation for Videos, a video compression framework based on hybrid neural representation with dual spatio-temporal feature modulation. The framework is designed to improve rate-distortion performance by effectively capturing the inherent spatio-temporal correlation of video data. Specifically, we introduce a temporal-aware affine transform layer, which leverages frame indices as conditions to align intermediate features with target frames. Furthermore, we incorporate a wavelet-based transform layer to modulate features in the spatial domain, significantly enhancing the network’s capability to recover high-frequency details. Experimental results demonstrate that our method outperforms Hybrid Neural Representation for Videos (HNeRV) and its upgraded version (HNeRV-Boost), improving the peak signal-to-noise ratio by +3.51 and +0.58 dB on the Bunny dataset, respectively.

Read at publisher →