AI Paper Atlas

Research map / Video / Video segmentation and tracking

DeRVOS: Decoupling Consistent Trajectory Generation and Multimodal Understanding for Referring Video Object Segmentation

Wenxuan Cheng, Ming Dai, Huimin Lu, Wankou Yang

CVPR 2026 · Video › Video segmentation and tracking · approach: Matching-based video object segmentation

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 Mask2Former with VITA then DVIS++ as frozen trajectory generator (ablation)DeRVOS · MeViS valuJ&F: 55.5 -> 56.0 -> 57.4Improved
replace BERT-B with ViLT-B then BEiT3-B as multimodal encoder (ablation)DeRVOS · MeViS valuJ&F: 55.9 -> 56.6 -> 57.4Improved
add CFMA instead of Text-Direct Integrator (ablation)DeRVOS · MeViS valuJ&F: 56.1 -> 56.5Improved
add motion-guided implicit selector on top of CFMA (full TAIS) (ablation)DeRVOS · MeViS valuJ&F: 56.5 -> 57.4Improved
use video-query integrator instead of text-direct integrator (ablation)DeRVOS · MeViS valuJ&F: 56.1 -> 56.2No clear effect
increase trajectory branch input resolution from 320 to 640 (ablation)DeRVOS · MeViS valuJ&F: 56.9 -> 60.9Improved
pretrain DVIS++ on OVIS rather than YouTube-VIS 2019, 2021 or VIPSeg (ablation)DeRVOS · MeViS valuJ&F: 55.4 -> 57.4 (vs YouTube-VIS 2019)Improved
add image-level pretraining on RefCOCO/+/gDeRVOS · MeViS valJ&F: 51.8 -> 56.0Improved
decoupled design with frozen trajectory generator versus DsHmpDeRVOS · MeViS valJ&F: 46.4 -> 51.8; training 20 hours -> 9 hoursImproved
DeRVOS vs DeRIS-B on RefCOCO image-level segmentationDeRVOS · RefCOCO valmIoU: 82.0 vs 80.2Hurt

Later papers that compared with this method

1 later accepted paper reports DeRVOS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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