AI Paper Atlas

Research map / Video / Video segmentation and tracking

Characterizing and Improving Stability in Neural Style Transfer

Agrim Gupta, Justin Johnson, Alexandre Alahi, Li Fei-Fei

ICCV 2017 · Video › Video segmentation and tracking · approach: Video restoration with temporal priors · 124 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
replace per-frame feedforward stylization with a recurrent network trained with optical-flow temporal consistency lossReal-Time baseline · DAVIS stability, Composition XIVSSIM between stylized adjacent patches: 0.31 -> 0.42Improved
add recurrent input and temporal consistency lossReal-Time baseline · DAVIS stability, MetzingerSSIM between stylized adjacent patches: 0.31 -> 0.42Improved
add recurrent input and temporal consistency lossReal-Time baseline · DAVIS stability, Rain PrincessSSIM between stylized adjacent patches: 0.41 -> 0.49Improved
add recurrent input and temporal consistency lossReal-Time baseline · DAVIS stability, The WavePSNR between stylized adjacent patches: 24.3 -> 24.8Improved
add recurrent input and temporal consistency lossReal-Time baseline · DAVIS stability, MosaicSSIM between stylized adjacent patches: 0.31 -> 0.39Improved
use feedforward recurrent network instead of per-frame optimizationOptim baseline · DAVIS stability, The WavePSNR between stylized adjacent patches: 25.5 -> 24.8Hurt
use feedforward recurrent network instead of per-frame optimizationOptim baseline · DAVIS stability, SSIM across five stylesSSIM between stylized adjacent patches: 0.48 -> 0.54Improved
use feedforward network instead of optimizationOptim baseline · Per-frame runtime, 1024 x 1024seconds per frame: 199.6 -> 0.141Improved
use feedforward network instead of optimizationOptim baseline · Per-frame runtime, 256 x 256seconds per frame: 22.14 -> 0.024Improved
add recurrent temporal-consistency training versus per-frame networkReal-Time baseline · Per-frame runtime, 512 x 512seconds per frame: 0.044 -> 0.044No clear effect
compare ours to Real-Time baseline in user studyReal-Time baseline · User study (AMT), More Flickervideos judged to flicker more: 26 vs 193Improved
compare ours to Real-Time baseline in user studyReal-Time baseline · User study (AMT), Overall Prefervideos preferred: 133 vs 80Improved

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