Research map / Video / Action recognition
Multiview Transformers for Video Recognition
Check 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 |
|---|---|---|---|
| Process multiple tubelet views with cross-view attention instead of one view (ablation) | MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.3 -> 81.1 (+2.8) | Helped |
| Add a second tubelet view (ablation) | B/4 on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.3 -> 80.8 (+2.5) | Helped |
| Use cross-view attention instead of late fusion (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 80.6 -> 81.1 | Helped |
| Use cross-view attention instead of bottleneck-token fusion (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 81.0 -> 81.1 | Helped |
| Replace cross-view attention with MLP fusion (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 81.1 -> 80.6 | Hurt |
| Replace integrated cross-view fusion with a separately trained ensemble (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.3 -> 77.7 | Hurt |
| Assign the larger encoder to the smaller-tubelet view (ablation) | Two-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.5 -> 81.8 | Helped |
| Increase a single-view Base encoder to 14 layers rather than add a view (ablation) | B/4 on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.3 -> 78.1 (-0.2) | Hurt |
| Increase a single-view Base encoder to 17 layers rather than add views (ablation) | B/4 on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 78.3 -> 78.4 (+0.1) | No clear effect |
| Use the same Base encoder across views (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 81.1 -> 81.1 | No clear effect |
| Apply cross-view attention at the middle and late layers (ablation) | Three-view MTV on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 80.96 -> 81.14 | Helped |
| Increase input resolution to 320p | MTV-B on Kinetics 400 · Kinetics 400 | Top-1 accuracy: 81.8 -> 82.4 | Helped |
Closest papers by meaning
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.85)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.85)
- Unifying Specialized Visual Encoders for Video Language Models (ICML 2025 · similarity 0.85)
- Understanding Video Transformers via Universal Concept Discovery (CVPR 2024 · similarity 0.85)
- Memory Consolidation Enables Long-Context Video Understanding (ICML 2024 · similarity 0.85)
- OneVision-Encoder: Codec-Aligned Sparsity as a Foundational Principle for Multimodal Intelligence (NeurIPS 2026 · 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.