Research map / Video / Video segmentation and tracking
PhysFormer: Facial Video-Based Physiological Measurement With Temporal Difference Transformer
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 PhysFormer rather than the strongest listed baseline | HR estimation · MMSE-HR cross-dataset | RMSE (bpm): 5.43 -> 5.36 | Helped |
| Use PhysFormer rather than the strongest listed baseline | HR estimation · OBF | RMSE (bpm): 1.26 -> 0.804 | Helped |
| Add the shallow stem (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 10.62 -> 7.56 | Helped |
| Use 4 × 4 × 4 tubes rather than full spatial projection with 4 × 16 × 16 tubes (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 10.61 -> 7.56 | Helped |
| Use smaller 2 × 4 × 4 temporal tubes (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 7.56 -> 7.81 | Hurt |
| Add temporal-difference multi-head self-attention (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 9.81 -> 7.56 | Helped |
| Use a smaller attention temperature τ=2.0 instead of τ≈4.9 (ablation) | PhysFormer TD-MHSA · VIPL-HR Fold-1 | RMSE (bpm): 9.51 -> 7.56 | Helped |
| Use temporal-difference MHSA rather than vanilla MHSA (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 10.43 -> 7.56 | Helped |
| Use spatio-temporal FF rather than vanilla FF (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 8.27 -> 7.56 | Helped |
| Use exponential dynamic frequency-loss scheduling rather than fixed or linear weighting (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 8.37 -> 7.56 | Helped |
| Add label distribution loss to frequency cross-entropy loss (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 8.09 -> 7.56 | Helped |
| Use ground-truth PPG PSD as the label distribution instead of the constructed distribution (ablation) | PhysFormer · VIPL-HR Fold-1 | RMSE (bpm): 8.21 -> 8.72 | Hurt |
Closest papers by meaning
- RhythmMamba: Fast, Lightweight, and Accurate Remote Physiological Measurement (AAAI 2025 · similarity 0.92)
- Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios (CVPR 2025 · similarity 0.92)
- PhysDiff: Physiology-based Dynamicity Disentangled Diffusion Model for Remote Physiological Measurement (AAAI 2025 · similarity 0.91)
- FacePhys: State of the Heart Learning (NeurIPS 2026 · similarity 0.90)
- FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing (NeurIPS 2024 · similarity 0.89)
- PaPaGei: Open Foundation Models for Optical Physiological Signals (ICLR 2025 · similarity 0.87)
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.