AI Paper Atlas

Research map / Image generation / GANs

Neural Video Compression Using GANs for Detail Synthesis and Propagation

Fabian Mentzer, Eirikur Agustsson, Johannes Ballé, David Minnen, Nick Johnston, George Toderici

ECCV 2022 · Image generation › GANs · approach: GAN latent optimization and evaluation · 50 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add GAN loss (ablation)Ours · MCL-JCV user studyuser preference / PSNR: PSNR 35.1 -> 34.5; preferred 68% of the timeImproved
add free latent from warped previous reconstruction (ablation)Ours · MCL-JCV user studyuser preference: 23% prefer No free latent over OursImproved
add conditioning of the discriminator on latents (ablation)Ours · MCL-JCV user studyuser preference: 28% prefer Uncond. Disc. over OursImproved
add UFlow supervised flow (ablation)Ours · MCL-JCV user studyuser preference: 33% prefer No UFlow over OursImproved
use bicubic resampling kernel in decoupled scale-space warping (ablation)MSE-trained DSSW model · MSE R-Dbpp at equal quality: R-D performance improves by 6%Improved
decouple scale-space warping into warp plus adaptive blurSSW · GPU runtimespeed: 2-3x fasterImproved
remove LPIPS distortion loss (use MSE only)Ours · trainingtraining speed: speeds up training by 35%Improved
raise bitrate of HEVC baseline to 0.14 and 0.22 bppHEVC · MCL-JCV user studyuser preference: 39% -> 41.2% -> 41.4% prefer HEVCMixed
condition discriminator on more framesOurs · MCL-JCVreconstruction qualityNo clear effect
replace GAN-trained model by perceptual metrics as predictor of studyLPIPS/PIM/FID · MCL-JCV user studymetric predicts study: predict neural codec studies, not H.264 or HEVC studiesMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy