Research map / Video / Video segmentation and tracking
Towards Global Video Scene Segmentation with Context-Aware Transformer
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 CAT with local-global self-attention and shot-to-scene pre-training instead of BaSSL | Video scene segmentation · MovieNet Test (M.S-318; pre-training on Train.+Test.+Val.) | AP: 57.40 -> 59.55 | Helped |
| Use CAT instead of BaSSL with pre-training on Train only | Video scene segmentation · MovieNet Test (M.S-318; pre-training on Train only) | AP: 53.36 -> 59.22 | Helped |
| Add local attention heads alongside global attention heads (ablation) | CAT · MovieNet Test (M.S-318) | AP: 58.35 -> 59.55 | Helped |
| Use only local attention heads instead of the Nl=2, Ng=6 local-global configuration (ablation) | CAT · MovieNet Test (M.S-318) | AP: 59.55 -> 58.92 | Hurt |
| Set the local encoder window to L=5 rather than L=3 (ablation) | CAT · MovieNet Test (M.S-318) | AP: 58.07 -> 59.55 | Helped |
| Increase the local encoder window from L=5 to L=7 (ablation) | CAT · MovieNet Test (M.S-318) | AP: 59.55 -> 59.49 | Hurt |
| Set the scene-level contextual window to P=17 rather than P=13 (ablation) | CAT · MovieNet Test (M.S-318) | AP: 58.73 -> 59.55 | Helped |
| Increase the scene-level contextual window from P=17 to P=19 (ablation) | CAT · MovieNet Test (M.S-318) | AP: 59.55 -> 59.14 | Hurt |
| Use Global Scene Matching as the sole pre-training task rather than Shot Masked Modeling (ablation) | CAT pre-training · MovieNet Test (M.S-318) | AP: 16.61 -> 44.27 | Helped |
| Use Shot Order Modeling as the sole pre-training task rather than Shot Masked Modeling (ablation) | CAT pre-training · MovieNet Test (M.S-318) | AP: 16.61 -> 39.59 | Helped |
| Use Local Scene Matching as the sole pre-training task rather than Shot Masked Modeling (ablation) | CAT pre-training · MovieNet Test (M.S-318) | AP: 16.61 -> 29.38 | Helped |
Closest papers by meaning
- Scene Consistency Representation Learning for Video Scene Segmentation (CVPR 2022 · similarity 0.90)
- Video Scene Segmentation with Genre and Duration Signals (ICLR 2026 · similarity 0.89)
- Efficient Temporal Action Segmentation via Boundary-aware Query Voting (NeurIPS 2024 · similarity 0.85)
- Neighbor Relations Matter in Video Scene Detection (CVPR 2024 · similarity 0.85)
- DyST: Towards Dynamic Neural Scene Representations on Real-World Videos (ICLR 2024 · similarity 0.85)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.84)
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.