AI Paper Atlas

Research map / Video / Video segmentation and tracking

Local-Global Context Aware Transformer for Language-Guided Video Segmentation

Chen Liang, Wenguan Wang, Tianfei Zhou, Jiaxu Miao, Yawei Luo, Yi Yang

TPAMI 2023 · Video › Video segmentation and tracking · approach: Matching-based video object segmentation · 116 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 cross-modality encoder to concatenation baseline (ablation)Baseline · A2D-S testmIoU: 47.3 -> 54.6Improved
add local-global memory (ablation)Locater with E_vw · A2D-S testmIoU: 54.6 -> 57.5Improved
add deeply-supervised learning on top of E_vw and memory (ablation)Locater · A2D-S testmIoU: 57.5 -> 59.7Improved
add memory on top of E_vw and DSL (ablation)Locater · A2D-S+ (avg)mIoU: 34.8 -> 36.4Improved
use Transformer visual encoder instead of I3D (ablation)Locater · A2D-S testmIoU: 58.3 -> 59.7Improved
add a decoder with product distance rather than encoder-only head (ablation)Locater · A2D-S testmIoU: 56.8 -> 59.7Improved
use product distance instead of cosine distance (ablation)Locater · A2D-S testmIoU: 59.2 -> 59.7Improved
use visual-guided query instead of text-invented query (ablation)Locater · A2D-S testmIoU: 58.7 -> 59.7Improved
stack cross-modality encoder modules K from 1 to 3 (ablation)Locater · A2D-S testmIoU: 58.4 -> 59.7Improved
increase cross-modality encoder modules K from 3 to 5 (ablation)Locater · A2D-S testmIoU: 59.7 -> 59.7No clear effect
sample frames with intervals of 3 to 10 for global memory instead of 30 (ablation)Locater · A2D-S testmIoU: 59.0 -> 59.7No clear effect
add local and global memory capacity (N_l=0,N_g=0 to 2N_v,1.5N_v) (ablation)Locater · A2D-S+ (avg)mIoU: 34.8 -> 36.4Improved

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