AI Paper Atlas

Research map / Video / Video segmentation and tracking

ViSt3D: Video Stylization with 3D CNN

Ayush Pande, Gaurav Sharma

NeurIPS 2023 · Video › Video segmentation and tracking · approach: Layered video decomposition and dynamics · 0 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 appearance subnetworks that disentangle appearance from motion before AdaIN3D (ablation)ViSt3D (C3D encoder + AdaIN3D) · video stylization, qualitativemotion artifactsImproved
add temporal loss (optical-flow based) (ablation)ViSt3D · Optical flow error, 20 test pairs (mean)mean abs optical flow difference: 1.63 -> 1.00Improved
add temporal loss (optical-flow based) (ablation)ViSt3D · video stylization, qualitativecolor consistency and motion preservationImproved
add intra-clip loss on top of temporal loss (ablation)ViSt3D+temporal · video stylization, qualitativeflashing color inconsistenciesImproved
add intra-clip loss on top of temporal loss (ablation)ViSt3D+temporal · Optical flow error, 20 test pairs (mean)mean abs optical flow difference: 1.00 -> 1.11Hurt
use static style clip with 3D encoder instead of 2D CNN feature statistics transferViSt3D encoder · initial experimentsstylization qualityImproved
compare ViSt3D against AdaAttN on content preservationViSt3D · User studycontent preservation preference (%): 41.0 -> 43.3Improved
compare ViSt3D against MCCNet on style degreeViSt3D · User studystyle degree preference (%): 31.3 vs 16.7Hurt
compare ViSt3D against AdaAttN on overall preferenceViSt3D · User studyoverall preference (%): 28.3 -> 28.7Improved
3D CNN instead of 2D AdaAttN on computeViSt3D · 144 frames 640x360processing time (s): 14 seconds -> 60 secondsHurt
compare ViSt3D (full) with AdaAttN on optical flow errorViSt3D+temporal+intra · Optical flow error, 20 test pairs (mean)mean abs optical flow difference: 1.26 vs 1.11Improved

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