Research map / Video / Video-language understanding
STAC: Selective Spatiotemporal Aggregation and Compression for Video Reasoning Segmentation
Hesham Syed, Yun Liu, Guolei Sun, Jing Yang, Henghui Ding, Xue Geng, Xudong Jiang
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 unidirectional temporal scan to spatial scan (ablation) | STAC SSA · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 44.1 -> 49.6 | Improved |
| add bidirectional spatial scan to temporal scan (ablation) | STAC SSA · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 45.2 -> 49.6 | Improved |
| use bidirectional temporal scan instead of unidirectional (ablation) | STAC SSA · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 42.7 vs 49.6 | Hurt |
| merge tokens by averaging instead of pruning (ablation) | STAC HSC · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 47.9 -> 49.6 | Improved |
| compress temporally first instead of spatially first (ablation) | STAC HSC · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 45.7 -> 49.6 | Improved |
| scan spatially first instead of temporally first (ablation) | STAC SSA · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 31.3 -> 45.7 | Improved |
| add SSA enhancement plus compression to uncompressed baseline (ablation) | STAC · ReasonVOS ablation (32 frames, 3K iterations) | J&F: 48.1 -> 49.6 | Improved |
| causal online inference instead of offline | STAC · Ref-DAVIS17 | J&F: 73.6 -> 72.5 | Hurt |
| causal online inference instead of offline | STAC · ReasonVOS | J&F: 52.3 -> 48.8 | Hurt |
| compress 85% of visual tokens with STAC vs GLUS full tokens | STAC · ReasonVOS | J&F: 49.8 -> 52.3 | Improved |
| compress 85% of visual tokens with STAC vs GLUS full tokens | STAC · MeViS | J&F: 50.3 -> 49.9 | Mixed |
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.
- VideoGLaMM — VideoGLaMM : A Large Multimodal Model for Pixel-Level Visual Grounding in Videos (CVPR 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
- The Devil is in Temporal Token: High Quality Video Reasoning Segmentation (CVPR 2025 · similarity 0.89)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.86)
- Fast Reasoning Segmentation for Images and Videos (CVPR 2026 · similarity 0.86)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.86)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.85)
- Tool-Augmented Spatiotemporal Reasoning for Streamlining Video Question Answering Task (NeurIPS 2025 · similarity 0.85)
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.
