Research map / Video / Video segmentation and tracking
Self-Supervised Representation Learning Framework for Remote Physiological Measurement Using Spatiotemporal Augmentation Loss
Hao Wang, Euijoon Ahn, Jinman Kim
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 |
|---|---|---|---|
| Use landmark-based facial spatial augmentation rather than standard spatial augmentations (ablation) | SLF-RPM · HR estimation | MAE (bpm): 9.54 -> 5.12 | Improved |
| Include all seven facial ROIs instead of omitting any one ROI (ablation) | SLF-RPM · HR estimation | MAE (bpm): 5.82 -> 5.12 | Improved |
| Use sparsity-based temporal augmentation rather than standard temporal transformations (ablation) | SLF-RPM · HR estimation | MAE (bpm): 5.74 -> 5.28 | Improved |
| Use sparsity range {1, 2, 3, 4, 5} rather than the smaller range {1, 2, 3, 4} (ablation) | SLF-RPM · HR estimation | MAE (bpm): 6.21 -> 5.28 | Improved |
| Extend the temporal sparsity range to include stride 6 (ablation) | SLF-RPM · HR estimation | MAE (bpm): 5.28 -> 5.92 | Hurt |
| Add pseudo-label auxiliary classification tasks to the contrastive objective (ablation) | SLF-RPM · MAHNOB-HCI | MAE (bpm): 4.25 -> 3.6 | Improved |
| Use simCLR rather than CPC, CPC + MoCo, or MoCo (ablation) | SLF-RPM · HR estimation | MAE (bpm): 13.49, 13.05, 11.73 -> 9.54 | Improved |
| Use SLF-RPM rather than the strongest SSL baseline, SeCo | Linear evaluation · MAHNOB-HCI | MAE (bpm): 7.03 -> 3.60 | Improved |
| Use SLF-RPM rather than the strongest SSL baseline, SeCo | Linear evaluation · VIPL-HR-V2 | MAE (bpm): 13.32 -> 12.56 | Improved |
| Use SLF-RPM rather than the strongest SSL baseline, SeCo | Linear evaluation · UBFC-rPPG | MAE (bpm): 9.83 -> 8.39 | Improved |
| Self-supervise pre-training on VIPL-HR-V2 before fine-tuning instead of training the supervised baseline | 3D ResNet-18 · MAHNOB-HCI | MAE (bpm): 7.34 -> 6.23 | Improved |
| Self-supervise pre-training on VIPL-HR-V2 before fine-tuning instead of training the supervised baseline | 3D ResNet-18 · UBFC-rPPG | MAE (bpm): 8.08 -> 7.35 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SeCo — SeCo: Exploring Sequence Supervision for Unsupervised Representation Learning (AAAI 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contrast-Phys+: Unsupervised and Weakly-Supervised Video-Based Remote Physiological Measurement via Spatiotemporal Contrast (TPAMI 2024 · similarity 0.91)
- The Way to My Heart Is Through Contrastive Learning: Remote Photoplethysmography From Unlabelled Video (ICCV 2021 · similarity 0.91)
- FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing (NeurIPS 2024 · similarity 0.87)
- PhysDiff: Physiology-based Dynamicity Disentangled Diffusion Model for Remote Physiological Measurement (AAAI 2025 · similarity 0.86)
- Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios (CVPR 2025 · similarity 0.86)
- RhythmMamba: Fast, Lightweight, and Accurate Remote Physiological Measurement (AAAI 2025 · similarity 0.86)
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.
