AI Paper Atlas

Research map / Video / Video segmentation and tracking

OneVOS: Unifying Video Object Segmentation with All-in-One Transformer Framework

Wanyun Li, Pinxue Guo, Xinyu Zhou, Lingyi Hong, Yangji He, Xiangyu Zheng, Wei Zhang, Wenqiang Zhang

ECCV 2024 · Video › Video segmentation and tracking · approach: Matching-based video object segmentation · 9 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 All-in-One modeling (joint extraction, matching, memory in one transformer) (ablation)discrete multi-object baseline with ConvViT · DAVIS 17 valJ&F: 82.8 -> 87.5Improved
add current-to-reference attention decoupling alone (ablation)All-in-One transformer · LVOS valJ&F: 64.9 -> 61.0Hurt
add memory-to-reference decoupling alone (ablation)All-in-One transformer · LVOS valJ&F: 64.9 -> 62.3Hurt
add both decouplings (Uni-Hybrid Attention) (ablation)All-in-One transformer · LVOS valJ&F: 64.9 -> 67.4Improved
add Dynamic Token Selector with dynamic memory (ablation)OneVOS · LVOS valJ&F: 67.4 -> 70.1Improved
add Dynamic Token Selector with dynamic memory (ablation)OneVOS · DAVIS 17 valJ&F: 88.5 -> 88.8Improved
force fixed 0.5 selection rate per layer in DTS (ablation)DTS · LVOS valJ&F: 70.1 -> 60.2Hurt
guide DTS with pooled current-frame tokens (ablation)DTS · LVOS valJ&F: 70.1 -> 66.4Hurt
add mask embedding at all layers instead of first 6 (ablation)OneVOS · MOSEJ&F: 60.9 -> 62.3Improved
increase memory capacity Cap from 1 to 4 (ablation)OneVOS · DAVIS 17 valJ&F: 87.5 -> 88.5 (FPS 16.6 -> 12.3)Improved
retain 1024 instead of 512 top-k memory tokens (ablation)OneVOS-D · LVOS valJ&F: 70.1 -> 68.7Hurt
use ConvViT-base instead of plain MAE ViT backbone (ablation)OneVOS · DAVIS 17 valJ&F: 87.5 -> 88.5Improved

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