Research map / Video / Video-language understanding
CAST: Modeling Visual State Transitions for Consistent Video Retrieval
Yanqing Liu, Yingcheng Liu, Fanghong Dong, Budianto Budianto, Cihang Xie, Yan jiao
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 |
|---|---|---|---|
| replace direct target prediction with additive residual update (ablation) | Early Fusion · CVR YouCook2 val | Accuracy: 35.99 -> 38.95 | Improved |
| replace direct target prediction with additive residual update (ablation) | Early Fusion · CVR YouCook2 val | State Accuracy: 38.92 -> 43.51 | Improved |
| use dual-path design with cross-attention instead of early concatenation (ablation) | CAST with residual · CVR YouCook2 val | Accuracy: +4.05% Acc. | Improved |
| add CAST prediction score C to text matching (semantic ensemble) (ablation) | Inference scoring · CVR YouCook2 val | Accuracy: 25.03 -> 45.46 | Improved |
| add visual continuity score B to semantic ensemble (ablation) | Full Ensemble · CVR YouCook2 val | Ident. Acc.: 70.38 -> 78.48 | Improved |
| add visual continuity score B to semantic ensemble (ablation) | Full Ensemble · CVR YouCook2 val | Accuracy: 45.46 -> 44.77 | Hurt |
| use CAST prediction alone vs text matching alone (ablation) | CAST score C · CVR YouCook2 val | Accuracy: 25.03 -> 42.60 | Improved |
| increase context length from L=0 to L=1 (ablation) | CAST · CVR YouCook2, COIN, CrossTask | Accuracy and Ident. Acc. | Improved |
| increase context length beyond L=1 (ablation) | CAST · CVR YouCook2, COIN, CrossTask | Accuracy | No clear effect |
| add CAST adapter to InternVideo2-1B | InternVideo2-1B · CVR YouCook2 | Accuracy: 36.75 -> 71.68 | Improved |
| add CAST adapter to InternVideo2-1B | InternVideo2-1B · CVR CrossTask | Accuracy: 20.61 -> 64.36 | Improved |
| add CAST adapter to Qwen3-VL-Embedding-2B | Qwen3-VL-Embedding-2B · CVR YouCook2 | Accuracy: 33.45 -> 56.64 | Improved |
Closest papers by meaning
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- From Play to Replay: Composed Video Retrieval for Temporally Fine-Grained Videos (NeurIPS 2025 · similarity 0.88)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.87)
- Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension (NeurIPS 2025 · similarity 0.87)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.87)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · 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.
