Research map / Video / Video segmentation and tracking
Video Instance Segmentation using Inter-Frame Communication Transformers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| use IFC memory-token communication instead of no communication (ablation) | IFC encoder · YouTube-VIS 2019 val T=5 | AP: 37.4 -> 39.0 | Helped |
| IFC vs full space-time attention (Full THW) (ablation) | encoder · YouTube-VIS 2019 val T=20 | AP: 39.7 -> 40.4 | Helped |
| use full THW attention instead of no communication (ablation) | encoder · YouTube-VIS 2019 val T=5 | AP: 37.4 -> 37.2 | No clear effect |
| decompose space-time attention (T-HW) instead of no communication (ablation) | encoder · YouTube-VIS 2019 val T=20 | AP: 39.6 -> 39.0 | Hurt |
| add memory tokens for image instance segmentation (ablation) | IFC · COCO val | AP: 35.0 -> 35.1 | No clear effect |
| use M=8 memory tokens instead of M=1 (ablation) | IFC · YouTube-VIS 2019 val T=20 | AP: 39.4 -> 40.4 | Helped |
| increase memory tokens from 8 to 16 (ablation) | IFC · YouTube-VIS 2019 val T=5 | AP: 39.0 -> 38.1 | Hurt |
| per-index memory decomposition instead of unified MT tokens (ablation) | IFC · YouTube-VIS 2019 val T=5 | AP: 38.1 -> 39.0 | Helped |
| mask-based instead of box-based bipartite matching (ablation) | IFC · YouTube-VIS 2019 val | AP: 37.2 -> 39.4 | Helped |
| increase clip stride from 1 to 3 (ablation) | IFC · YouTube-VIS 2019 val | AP / FPS: approximately 150% speed improvement with only 0.1% AP drop | No clear effect |
| use IFC instead of VisTR (ResNet-101) | video instance segmentation · YouTube-VIS 2019 val | AP: 40.1 -> 44.6 | Helped |
Closest papers by meaning
- End-to-End Video Instance Segmentation With Transformers (CVPR 2021 · similarity 0.91)
- Decoupled Motion Expression Video Segmentation (CVPR 2025 · similarity 0.87)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.86)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.86)
- VideoCutLER: Surprisingly Simple Unsupervised Video Instance Segmentation (CVPR 2024 · similarity 0.86)
- Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · similarity 0.86)
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.