List of participants of MSU Super-Resolution for Video Compression Benchmark 2021

---

DBVSR

Estimates a motion blur for the particular input. Compensates the motion between frames explicitly.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

DynaVSR

DynaVSR can be applied to any VSR deep-learning model. For our benchmark, we used pretrained weights for model EDVR, which uses Deformable convolution to align neighboring frames.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

EGVSR

The generator part is divided into FNet module and SRNet module for optical flow estimation and video frame super-resolution, respectively.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

iSeeBetter

A combination of an RNN-based optical flow method that preserves spatio-temporal information in the current and adjacent frames as the generator and a discriminator that is adept at ensuring the generated SR frame offers superior fidelity.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

LGFN

Uses deformable convolutions with decreased multi-dilation convolution units (DMDCUs) to align frames explicitly. Fuses features from local and global fusion modules.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

RealSR

Tries to estimate degradation kernel and noise distribution for better visual quality.

Added to the benchmark by MSU G&M Lab

Links: GitHub, GitHub, paper

SOF-VSR

Two models: SOF-VSR-BD (trained on gauss degradation type), SOF-VSR-BI (trained on bicubic degradation type)

Compensates motion by high-resolution optical flow, estimated from the low-resolution one in a coarse-to-fine manner.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

Topaz Video Enhance AI

Topaz Video Enhance AI is a commercial filter.

Three models:
• ahq-11 (Artemis High Quality v11) - upscale or sharpen high quality input video, reducing motion flicker,
• amq-12 (Artemis Medium Quality v12) - upscale or enhance medium quality video with moderate noise or compression artifacts,
• amqs-1 (Artemis Dehalo v1) - upscale or enhance medium quality progressive video that contains haloing, moderate noise or compression artifacts.

Added to the benchmark by MSU G&M Lab

Links: Website

waifu2x-ncnn-vulkan

Two models: waifu2x-anime and waifu2x-cunet

NCNN implementation of waifu2x converter. Waifu2x-ncnn-vulkan uses NCNN project as the universal neural network inference framework.

Added to the benchmark by MSU G&M Lab

Links: GitHub, GitHub