Skip to content
/ LGFN Public

Local-Global Fusion Network for Video Super-Resolution

License

Notifications You must be signed in to change notification settings

BIOINSu/LGFN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Local-Global Fusion Network for Video Super-Resolution

This repository is about Local-Global Fusioin Network for Video Super-Resolution (LGFN).

Introduction

Our proposed Local-Global Fusion Network for Video Super-Resolution (LGFN) devotes to effectively addressing the problem of restoring low-resolution (LR) videos to high-resolution (HR) ones. Features are extracted from the input frames through a feature extraction module. Then, we align the neighboring frames with the reference frame through stacked deformable convolutions (DCs) with decreased multi-dilation convolution units (DMDCUs) to predict more accurate parameters. After that, features are fused by local fusion module and global fusion module respectively. The outputs are integrated together and sent into the reconstruction module to restore high resolution video frames.

Dependencies

Dataset

  • Vimeo_90k (extracted code: "hjle")
  • Vid4 (extracted code: "htvb")
  • SPMCS (extracted code: "ouyg")

Pretrained Model

Experimental Results

Quantitative Results

Quantitative results of state-of-the-art SR algorithms on Vid4 for 4 ×. Bold type indicates the best and underline indicates the second best performance (PSNR/SSIM).

quantitative_vid4

Quantitative results of state-of-the-art SR algorithms on SPMCS for 4 ×. The results are the average evaluation across all restored video frames. Bold type indicates the best and underline indicates the second best performance (PSNR/SSIM).

quantitative_spmcs_all

Quantitative results of state-of-the-art SR algorithms on part of SPMCS clips for 4 ×. Bold type indicates the best and underline indicates the second best performance (PSNR/SSIM).

quantitative_spmcs_part

Quantitative results of state-of-the-art SR algorithms on Vimeo-90K-T for 4 ×. Bold type indicates the best and underline indicates the second best performance (PSNR/SSIM).

quantitative_vimeo90k

Visual Results

Visual comparisons on Vid4 for 4 × scaling factor. Zoom in to see better visualization.

visual_spmcs

Visual comparisons on SPMCS for 4 × scaling factor.

visual_spmcs

Visual comparisons on Vimeo-90K-T for 4 × scaling factor.

visual_vimeo90k

About

Local-Global Fusion Network for Video Super-Resolution

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published