Research map / Video / Video segmentation and tracking
Switch-a-View: View Selection Learned from Unlabeled In-the-wild Videos
Sagnik Majumder, Tushar Nagarajan, Ziad Al-Halah, Kristen Grauman
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 view-switch pretraining on unlabeled HowTo100M (ablation) | view selector · Ego-Exo4D view selection | AUC: 51.6 -> 57.3 | Improved |
| add shot-level pseudo-labeling instead of clip-level (ablation) | view-switch detector · HowTo100M view-switch | AUC: 51.9 -> 63.8 | Improved |
| add past narrations to (F, N') inputs (ablation) | view-switch detector · HowTo100M view-switch | AUC: 60.1 -> 63.8 | Improved |
| add next narration to (F, N) inputs (ablation) | view-switch detector · HowTo100M view-switch | AUC: 60.9 -> 63.8 | Improved |
| add past frames to N'-only input (ablation) | view-switch detector · HowTo100M view-switch | AUC: 57.8 -> 60.1 | Improved |
| use next narration instead of past narration (ablation) | view-switch detector · HowTo100M view-switch | AUC: 54.4 -> 57.8 | Improved |
| past frame context 8 s instead of 32 s (ablation) | view-switch detector · HowTo100M view-switch | AUC: 55.0 -> 63.8 | Improved |
| past frame context 8 s instead of 2 s (ablation) | view-switch detector · HowTo100M view-switch | AUC: 63.1 -> 63.8 | Improved |
| past narration context 32 s instead of 4 s (ablation) | view-switch detector · HowTo100M view-switch | AUC: 53.9 -> 63.8 | Improved |
| Switch-a-view vs best retrieval baseline | Switch-a-view · HowTo100M view-switch | AUC: 53.4 -> 63.8 | Improved |
| Switch-a-view zero-shot vs Retrieval-F | Switch-a-view · Ego-Exo4D view-switch | accuracy: 52.6 -> 51.2 | Hurt |
| Switch-a-view vs LangView-bigData with 98x more pretraining data | Switch-a-view · Ego-Exo4D view selection | AUC: 54.8 -> 57.3 | Improved |
Closest papers by meaning
- Which Viewpoint Shows it Best? Language for Weakly Supervising View Selection in Multi-view Instructional Videos (CVPR 2025 · similarity 0.86)
- Bootstrap Your Own Views: Masked Ego-Exo Modeling for Fine-grained View-invariant Video Representations (CVPR 2025 · similarity 0.86)
- Viewpoint Rosetta Stone: Unlocking Unpaired Ego-Exo Videos for View-invariant Representation Learning (CVPR 2025 · similarity 0.85)
- Step Differences in Instructional Video (CVPR 2024 · similarity 0.84)
- EgoExoBench: A Benchmark for First- and Third-person View Video Understanding in MLLMs (NeurIPS 2025 · similarity 0.84)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 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.
