Research map / Video / Video segmentation and tracking
Self-Supervised Motion Magnification by Backpropagating Through Optical Flow
Zhaoying Pan, Daniel Geng, Andrew Owens
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 self-supervised flow-based model instead of Oh et al. | Oh et al. · Real-world, alpha=10, GMFlow | motion error: 4.34 -> 3.88 | Improved |
| use self-supervised model instead of forward warping | Warp Bilinear · Real-world, alpha=10, PWC-Net | motion error: 4.12 -> 3.32 | Improved |
| use our model instead of FLAVR | FLAVR · Real-world, alpha=10, PWC-Net | motion error: 4.24 -> 3.32 | Improved |
| train with unsupervised ARFlow instead of supervised RAFT flow (ablation) | Ours · Real-world, alpha=10, PWC-Net | motion error: 3.32 -> 3.70 | Hurt |
| train with unsupervised ARFlow instead of supervised RAFT flow (ablation) | Ours · Synthetic noise test 100x | motion error: 164.36 -> 167.77 | Hurt |
| use our model instead of Oh et al. | Oh et al. · Synthetic noise test 100x | motion error: 181.26 -> 164.36 | Improved |
| use our model instead of Oh et al. | Oh et al. · Synthetic subpixel test 0.04px | motion error: 9.82 -> 10.02 | Hurt |
| use our model instead of Oh et al. | Oh et al. · Synthetic noise test 1x | SSIM: 0.97 -> 0.94 | Hurt |
| use our model instead of Oh et al. | Oh et al. · Synthetic noise test 100x | SSIM: 0.31 -> 0.10 | Hurt |
| use our model instead of Oh et al. | Oh et al. · Synthetic subpixel test 0.04px | magnification error: 193.78 -> 175.45 | Improved |
| use our model instead of Warp Bilinear at large magnification | Warp Bilinear · Real-world, alpha=64, RAFT evaluation | motion error: 28.70 -> 29.26 | Hurt |
| add test-time adaptation (finetune on input video) | Ours · Out-of-domain real videos | motion error | Improved |
Closest papers by meaning
- Learning-based Video Motion Magnification (ECCV 2018 · similarity 0.90)
- Event-Based Motion Magnification (ECCV 2024 · similarity 0.89)
- Video Acceleration Magnification (CVPR 2017 · similarity 0.89)
- Frequency Decoupling for Motion Magnification via Multi-Level Isomorphic Architecture (CVPR 2024 · similarity 0.88)
- EulerMormer: Robust Eulerian Motion Magnification via Dynamic Filtering within Transformer (AAAI 2024 · similarity 0.87)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 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.
