Research map / Video / Action recognition
Anticipating Daily Intention Using On-Wrist Motion Triggered Sensing
Tz-Ying Wu, Ting-An Chien, Cheng-Sheng Chan, Chan-Wei Hu, Min Sun
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 motion-triggered sensing (policy network) instead of always processing object images | RNN intention anticipation · Daily intention dataset, user A, 100% observed | accuracy: 93.29% -> 92.68% | Mixed |
| replace concatenating object and motion on every frame with motion-triggered sensing | RNN · Daily intention dataset, user B | triggered ratio / accuracy: 95.12% -> 90.85% | Mixed |
| add object representation to motion-only RNN (ablation) | RNN · Daily intention dataset, user A, 100% observed | accuracy: 75.61% -> 94.51% | Improved |
| add motion representation to object-only RNN (concatenation) (ablation) | RNN · Daily intention dataset, user A, 100% observed | accuracy: 94.51% -> 93.29% | No clear effect |
| add motion-triggering to motion-only sensing (ablation) | RNN · Daily intention dataset, user C, 25% observed | accuracy: 71.34% -> 94.51% | Improved |
| increase the ratio of triggered moments (ablation) | motion-triggered RNN · Daily intention dataset, user A | accuracy: accuracy only decreases slightly when ratio is larger than 20% | Improved |
| use 3 input channels with 3 layers instead of 1 layer (ablation) | hand motion 1D-CNN · Hand motion classification | testing accuracy: 77.21% -> 83.92% | Improved |
| use 3 input channels with 3 layers instead of 1ch-3layer prior model (ablation) | hand motion 1D-CNN · Hand motion classification | testing accuracy: 81.41% -> 83.92% | Improved |
| flip left-hand samples by negating one channel (ablation) | hand motion 1D-CNN · Left-hand motion classification | testing accuracy: 52.78% -> 83.33% | Improved |
| use ResNet-152 instead of ResNet-50 for object classification (ablation) | object CNN · Object classification | testing accuracy: 80.77% -> 83.09% | Improved |
| use ResNet-152 instead of ResNet-50 (ablation) | object CNN · Object classification on TX1 | speed: 6 fps -> 3 fps | Hurt |
Closest papers by meaning
- Time-Conditioned Action Anticipation in One Shot (CVPR 2019 · similarity 0.85)
- Bidirectional Progressive Transformer for Interaction Intention Anticipation (ECCV 2024 · similarity 0.83)
- Forecasting Human-Object Interaction: Joint Prediction of Motor Attention and Actions in First Person Video (ECCV 2020 · similarity 0.83)
- Intend to Move: A Multimodal Dataset for Intention-Aware Human Motion Understanding (NeurIPS 2025 · similarity 0.83)
- Intention-Guided Cognitive Reasoning for Egocentric Long-Term Action Anticipation (AAAI 2026 · similarity 0.83)
- Forecasting Action Through Contact Representations From First Person Video (TPAMI 2021 · 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.
