Research map / Video / Action recognition
The Wisdom of Crowds: Temporal Progressive Attention for Early Action Prediction
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 progressive fine-to-coarse temporal scales instead of equally sized segments (ablation) | TemPr · UCF-101 | accuracy: 83.7 -> 90.2 | Helped |
| use progressive scales rather than random-length scales (ablation) | TemPr · UCF-101 | accuracy: 88.8 -> 90.2 | Helped |
| combine predictor agreement and confidence with adaptive aggregation rather than averaging (ablation) | TemPr · UCF-101 | accuracy: 89.5 -> 90.2 | Helped |
| share classifier weights across attention towers (ablation) | TemPr · UCF-101 | accuracy: 84.7 -> 90.2 | Helped |
| share the Cross MAB latent array across towers (ablation) | TemPr · UCF-101 | accuracy: 90.2 -> 90.2 | No clear effect |
| use cross-attention MAB instead of self-attention MAB (ablation) | TemPr · UCF-101 | accuracy: 83.2 -> 90.2 | Helped |
| use 8 rather than 4 self-attention blocks (ablation) | TemPr · UCF-101 | accuracy: 83.4 -> 90.2 | Helped |
| use attention towers rather than MLP-only towers (ablation) | TemPr · UCF-101 | accuracy: 72.4 -> 90.2 | Helped |
| increase bottleneck dimension from 128 to 256 (ablation) | TemPr · UCF-101 | accuracy: 89.1 -> 90.2 | Helped |
| increase bottleneck dimension from 256 to 512 (ablation) | TemPr · UCF-101 | accuracy: 90.2 -> 90.7 | Helped |
| aggregate progressive scales instead of using only the final tower (ablation) | TemPr · UCF-101 | accuracy: 78.5 -> 84.3 | Helped |
| add progressive video scales and attention towers rather than predicting directly from CNN features (ablation) | TemPr · UCF-101 | accuracy: 69.4 -> 90.2 | Helped |
Closest papers by meaning
- Action-Guided Attention for Video Action Anticipation (ICLR 2026 · similarity 0.89)
- Progressive Teacher-Student Learning for Early Action Prediction (CVPR 2019 · similarity 0.87)
- Prediction-Feedback DETR for Temporal Action Detection (AAAI 2025 · similarity 0.86)
- ActFusion: a Unified Diffusion Model for Action Segmentation and Anticipation (NeurIPS 2024 · similarity 0.86)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.86)
- Can't Make an Omelette Without Breaking Some Eggs: Plausible Action Anticipation Using Large Video-Language Models (CVPR 2024 · 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.