AI Paper Atlas

Research map / Video / Video segmentation and tracking

Scalable Video Object Segmentation with Simplified Framework

Qiangqiang Wu, Tianyu Yang, Wei Wu, Antoni B. Chan

ICCV 2023 · Video › Video segmentation and tracking · approach: Matching-based video object segmentation · 46 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
initialize ViT with MAE pretraining instead of ImageNet1K supervised (ablation)SimVOS-B · DAVIS-2017 valJ&F: 81.3 -> 88.0Improved
use pretraining instead of random initialization (ablation)SimVOS-B · DAVIS-2017 valJ&F: 68.8 -> 81.3 (ImageNet1K), 88.0 (MAE)Improved
MAE instead of contrastive MoCo-V3 pretraining (ablation)SimVOS-B · DAVIS-2017 valJ&F: 81.5 -> 88.0Improved
scale backbone from ViT-Base to ViT-Large (ablation)SimVOS · DAVIS-2017 valJ&F: 88.0 -> 88.5 (FPS 4.9 -> 2.0)Improved
apply within-frame attention in more early layers (ablation)SimVOS-B · DAVIS-2017 valJ&F: 88.0 -> 86.5 at L=8; FPS 4.9 -> 8.8Mixed
add token refinement at L=4 (ablation)SimVOS-B · DAVIS-2017 valJ&F: 86.9 -> 87.1; FPS 6.5 -> 9.9No clear effect
add token refinement at L=2 (ablation)SimVOS-B · DAVIS-2017 valJ&F: 87.4% -> 86.0%Hurt
reduce prototypes from 384 to 128 each (ablation)SimVOS-BS · DAVIS-2017 valJ&F: 87.1% -> 85.3%Hurt
use balanced foreground/background prototype ratio (384/384) vs 2:1 or 1:2 (ablation)SimVOS-BS · DAVIS-2017 valJ&F: 87.1% vs 86.0% / 86.6%Improved
add mask-guided post-processing sigma in prototype generation (ablation)SimVOS-BS · DAVIS-2017 valJ&F: 84.9 -> 87.1Improved
joint feature extraction and matching in a single ViT without synthetic pretraining or BL30KSimVOS-B vs XMEM* · DAVIS-2017 valJ&F: 87.7 -> 88.0Improved
scale to SimVOS-L on harder test setSimVOS · DAVIS-2017 test-devJ&F: 80.4 -> 82.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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