This is the official implementation of the paper "DNeRV: Modeling Inherent Dynamics via Difference Neural Representation for Videos (CVPR2023)" and "PNeRV: Enhancing Spatial Consistency via Pyramidal Neural Representation for Videos (CVPR2024)".
Our work is modified base on "E-NeRV: Expedite Neural Video Representation with Disentangled Spatial-Temporal Context (ECCV 2022)", https://github.com/kyleleey/E-NeRV. Thanks to Zizhang Li for his excellent implementation. The code structure of the DNeRV is the same as E-NeRV.
Also, DNeRV is inspire by "HNeRV: A Hybrid Neural Representation for Videos (CVPR2023)". More than content stream proposed by HNeRV, we add the difference stream with a difference encoder and a attention module, Collaborative Content Unit (CCU).
use command: bash scripts/run.sh cfgs/DNeRV.yaml xxx 29500 to run the code.