Research map / Probabilistic and scientific modeling / Molecules and proteins
Uncertainty-Guided Probabilistic Transformer for Complex Action 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 |
|---|---|---|---|
| use UGPT instead of GHRM with the I3D backbone | Complex action recognition · Breakfast Actions | Activities accuracy (%): 75.49 -> 77.79 | Helped |
| use UGPT instead of GHRM with the I3D backbone | Complex action recognition · Breakfast Actions | Actions mAP (%): 65.86 -> 67.82 | Helped |
| use UGPT instead of GHRM with the I3D backbone | Complex action recognition · MultiTHUMOS | mAP (%): 79.89 -> 81.42 | Helped |
| add UGPT to ResNet-152 | Charades · Charades | mAP (%): 31.6 -> 35.7 | Helped |
| add UGPT to I3D | Charades · Charades | mAP (%): 38.3 -> 38.8 | Helped |
| add UGPT to 3D ResNet-101 | Charades · Charades | mAP (%): 41.1 -> 42.4 | Helped |
| replace deterministic Transformer with UGPT | Complex action recognition · Breakfast Actions, 20% training data | accuracy (%): 37.09 -> 46.70 | Helped |
| remove atomic action localization (ablation) | UGPT · Breakfast Actions | accuracy (%): 77.79 -> 78.56 | Helped |
| fine-tune atomic action localization jointly (ablation) | UGPT · Breakfast Actions atomic-action localization | accuracy (%): 70.2% to 70.4% | Helped |
| set the number of probabilistic self-attention layers to 6 rather than fewer tested layers (ablation) | UGPT · Breakfast Actions | performance | Helped |
| set NLL loss weight w to 0.15 rather than 0 (ablation) | UGPT · Breakfast Actions | accuracy (%): 76.69 -> 77.79 | Helped |
| increase NLL loss weight to 0.20 (ablation) | UGPT · Breakfast Actions | accuracy (%): 77.79 -> 70.52 | Hurt |
Closest papers by meaning
- ExACT: Language-guided Conceptual Reasoning and Uncertainty Estimation for Event-based Action Recognition and More (CVPR 2024 · similarity 0.85)
- Uncertainty-aware Action Decoupling Transformer for Action Anticipation (CVPR 2024 · similarity 0.83)
- LLaVAction: evaluating and training multi-modal large language models for action understanding (ICLR 2026 · similarity 0.83)
- VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention (AAAI 2025 · similarity 0.82)
- TrAction: Action Recognition with Sparse Trajectories (NeurIPS 2026 · similarity 0.82)
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.82)
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.