Research map / Video / Video segmentation and tracking
Entry-Flipped Transformer for Inference and Prediction of Participant Behavior
Bo Hu, Tat-Jen Cham
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 entry-flipped cross-attention in decoder (ablation) | Transformer · Tennis (prediction) | MAD Avg: 31.33 -> 28.44 | Improved |
| add entry-flipped cross-attention in decoder | Transformer · Tennis (inference) | FAD Avg: 49.34 -> 46.43 | Improved |
| flip with all past frames as query rather than limited query (ablation) | EF-Transformer · Tennis (prediction) | FAD Avg: 46.83 -> 51.71 | Hurt |
| use hybrid typical/EF decoders instead of pure EF (ablation) | EF-Transformer · Tennis (prediction) | MAD Avg: 28.44 -> 29.29 | Hurt |
| predict relative-to-first-frame uv^R instead of absolute uv (ablation) | Sequential ST-encoder · Tennis (inference) | MAD Avg: 48.89 -> 28.35 | Improved |
| predict cumulative frame differences instead of uv^R (ablation) | Sequential ST-encoder · Tennis (inference) | FAD Avg: 46.43 -> 54.93 | Hurt |
| use sequential ST-encoder instead of parallel S+T (ablation) | EF-Transformer · Tennis (inference) | MAD Avg: 28.93 -> 28.35 | Improved |
| add entry flipping under noise spike at t=9 | Transformer · Tennis noise spike at t=9 | FAD Avg: 157.81 -> 124.29 | Improved |
| add entry flipping under noise spike at t=3 | Transformer · Tennis noise spike at t=3 | FAD Avg: 99.67 -> 54.15 | Improved |
| add entry flipping in multi-task prediction with estimated actions | Transformer · Tennis multi-task prediction | FAD Avg: 52.01 -> 45.45 | Improved |
| add entry flipping on dance prediction (1 target) | Transformer · Dance | MAD Avg: 12.35 -> 9.39 | Improved |
| add entry flipping on NBA prediction | Transformer · NBA | MAD Avg: 3.99 -> 3.89 | Improved |
Closest papers by meaning
- Proxy-Bridged Game Transformer for Interactive Extreme Motion Prediction (ICCV 2025 · similarity 0.85)
- Social Diffusion: Long-term Multiple Human Motion Anticipation (ICCV 2023 · similarity 0.84)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · similarity 0.84)
- Behavior Transformers: Cloning $k$ modes with one stone (NeurIPS 2022 · similarity 0.84)
- Sports-Traj: A Unified Trajectory Generation Model for Multi-Agent Movement in Sports (ICLR 2025 · similarity 0.83)
- Spot The Ball: A Benchmark for Visual Social Inference (CVPR 2026 · similarity 0.83)
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.
