Research map / Video / Video segmentation and tracking
Multi-Task Temporal Shift Attention Networks for On-Device Contactless Vitals Measurement
Xin Liu, Josh Fromm, Shwetak Patel, Daniel McDuff
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 temporal modeling with temporal shift, hybrid 2D/3D, or 3D convolutions to 2D-CAN (ablation) | CAN · AFRL (All Tasks) | heart rate MAE: 2.32 -> 1.12 | Improved |
| Add temporal modeling with temporal shift, hybrid 2D/3D, or 3D convolutions to 2D-CAN (ablation) | CAN · AFRL (All Tasks) | respiration rate MAE: 2.86 -> 2.06 | Improved |
| Replace hybrid 2D/3D convolutions with temporal shift (ablation) | Hybrid-CAN · AFRL (All Tasks) | inference time (ms): 26 -> 12 | Improved |
| Replace hybrid 2D/3D convolutions with temporal shift (ablation) | Hybrid-CAN · AFRL (All Tasks) | heart rate MAE: 1.12 -> 1.32 | Hurt |
| Share the temporal-shift network between pulse and respiration tasks (ablation) | TS-CAN · AFRL (All Tasks) | heart rate MAE: 1.32 -> 1.45 | No clear effect |
| Share the temporal-shift network between pulse and respiration tasks (ablation) | TS-CAN · AFRL (All Tasks) | inference time (ms): 12 -> 6 | Improved |
| Use temporal shift instead of 2D-CAN temporal modeling (ablation) | 2D-CAN · AFRL (All Tasks) | heart rate MAE: 2.32 -> 1.32 | Improved |
| Use temporal shift instead of 2D-CAN temporal modeling (ablation) | 2D-CAN · AFRL (All Tasks) | inference time (ms): 20 -> 12 | Improved |
| Use the multi-task temporal-shift model rather than separate task models (ablation) | TS-CAN · AFRL (All Tasks) | heart rate MAE: 1.32 -> 1.45 | Hurt |
| Use 3D convolutions rather than temporal shift (ablation) | TS-CAN · AFRL (All Tasks) | inference time (ms): 12 -> 48 | Hurt |
Closest papers by meaning
- CogPhys: Assessing Cognitive Load via Multimodal Remote and Contact-based Physiological Sensing (NeurIPS 2025 · similarity 0.85)
- FacePhys: State of the Heart Learning (NeurIPS 2026 · similarity 0.85)
- PHASE-Net: Physics-Grounded Harmonic Attention System for Efficient Remote Photoplethysmography Measurement (CVPR 2026 · similarity 0.84)
- Fusion-Vital: Video-RF Fusion Transformer for Advanced Remote Physiological Measurement (AAAI 2024 · similarity 0.82)
- FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing (NeurIPS 2024 · similarity 0.82)
- RhythmMamba: Fast, Lightweight, and Accurate Remote Physiological Measurement (AAAI 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.
