Research map / Video / Video segmentation and tracking
QST-SAM: Leveraging Cross-modal Instructions for Few-shot Referring Video Object Segmentation
Xin Liu, Weijia Li, Tao Chen, Hongsong Wang, Guosen Xie, Caifeng Shan, Fang Zhao
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Q-S-T Transformer (with Support Instruction Compressor) (ablation) | QST-SAM · Ref-YouTube-VOS | J&F: 62.4 vs 60.8 (-1.6) | Improved |
| add Multi-Priors Prompting (ablation) | QST-SAM · Ref-YouTube-VOS | J&F: 62.4 vs 61.7 (-0.7) | Improved |
| add both Q-S-T Transformer and Multi-Priors Prompting (ablation) | QST-SAM · Ref-YouTube-VOS | J&F: 62.4 vs 60.4 (-2.0) | Improved |
| increase support instructions K from 4 to 32 (ablation) | QST-SAM · Ref-YouTube-VOS | J&F: 59.8 -> 62.4 | Improved |
| increase support instructions K from 32 to 64 (ablation) | QST-SAM · Ref-YouTube-VOS | J&F: 62.4 -> 62.6 (+0.2), memory 42.2 -> 47.6 GB | No clear effect |
| add 5-shot support images to SAMWISE with simple cross-attention fusion | SAMWISE · Ref-DAVIS17 | J&F: 65.5 -> 67.5 (+2.0) | Improved |
| add 5-shot support images to SAMWISE with simple cross-attention fusion | SAMWISE · Ref-YouTube-VOS | J&F: 59.3 -> 60.4 (+1.1) | Improved |
| replace SAMWISE simple fusion with QST-SAM | QST-SAM · Ref-DAVIS17 | J&F: 67.5 -> 70.3 | Improved |
| replace SAMWISE simple fusion with QST-SAM | QST-SAM · Ref-YouTube-VOS | J&F: 60.4 -> 62.4 | Improved |
| drop support-side text annotations (QST-SAM vs CMA-ISM) | QST-SAM · Mini-Ref-YouTube-VOS (legacy FSRVOS) | J&F: 54.0 -> 59.2 (+5.2) | Improved |
| add 5-shot support images to ReferDINO | ReferDINO · Ref-YouTube-VOS | J&F: 59.8 -> 60.6 (+0.8) | Improved |
| add 5-shot support images to LoSh | LoSh · Ref-DAVIS17 | J&F: 60.4 -> 64.9 (+4.5) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SAMWISE — SAMWISE: Infusing Wisdom in SAM2 for Text-Driven Video Segmentation (CVPR 2025) · better on 2 of 2 shared results
- ReferDINO — ReferDINO: Referring Video Object Segmentation with Visual Grounding Foundations (ICCV 2025) · better on 2 of 2 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
- Few-Shot Referring Video Single- and Multi-Object Segmentation Via Cross-Modal Affinity with Instance Sequence Matching (IJCV 2025 · similarity 0.91)
- End-to-End Referring Video Object Segmentation With Multimodal Transformers (CVPR 2022 · similarity 0.91)
- MPG-SAM 2: Adapting SAM 2 with Mask Priors and Global Context for Referring Video Object Segmentation (ICCV 2025 · similarity 0.90)
- Learning Cross-Modal Affinity for Referring Video Object Segmentation Targeting Limited Samples (ICCV 2023 · similarity 0.90)
- Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · similarity 0.90)
- SAMWISE: Infusing Wisdom in SAM2 for Text-Driven Video Segmentation (CVPR 2025 · similarity 0.89)
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.
