Zizhang Li, Mengmeng Wang, Huaijin Pi, Kechun Xu, Jianbiao Mei, Yong Liu
This is the official implementation of the paper "E-NeRV: Expedite Neural Video Representation with Disentangled Spatial-Temporal Context".
Recently, the image-wise implicit neural representation of videos, NeRV, has gained popularity for its promising results and swift speed compared to regular pixel-wise implicit representations. However, the redundant parameters within the network structure can cause a large model size when scaling up for desirable performance. The key reason of this phenomenon is the coupled formulation of NeRV, which outputs the spatial and temporal information of video frames directly from the frame index input. In this paper, we propose E-NeRV, which dramatically expedites NeRV by decomposing the image-wise implicit neural representation into separate spatial and temporal context. Under the guidance of this new formulation, our model greatly reduces the redundant model parameters, while retaining the representation ability. We experimentally find that our method can improve the performance to a large extent with fewer parameters, resulting in a more than 8$\times$ faster speed on convergence
We run with Python 3.8, you can set up a conda environment with all dependencies like so:
conda create -n enerv python=3.8
conda activate enerv
pip install -r requirements.txt The paper's main experiments were conducted on:
- data/bunny directory contains big buck bunny video sequence
- UVG video sequences can be found in official website
For now, we provide training and evaluation on original NeRV and our E-NeRV, by running following command with different configs and specified EXP_NAME:
bash scripts/run.sh ./cfgs/E-NeRV-bunny.yaml EXP_NAME 29500The logs and other outputs will be automaically saved to the output directory ./outputs/EXP_NAME
Thanks to Hao Chen for his excellent work and implementation NeRV.
If you have any questions, please feel free to email the authors.