Research map / Optimization / Implicit bias and generalization
Cost Function Unrolling in Unsupervised Optical Flow
Gal Lifshitz, Dan Raviv
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 |
|---|---|---|---|
| replace standard TV with unrolled cost | ARFlow · Sintel Clean | AEPE: 2.29 -> 2.17 | Improved |
| replace standard TV with unrolled cost | ARFlow · KITTI 2015 | AEPE: 2.91 -> 2.82 | Improved |
| replace standard TV with unrolled cost | SMURF · KITTI 2015 | AEPE: 3.17 -> 2.89 | Improved |
| replace standard TV with unrolled cost | ARFlow · Sintel Clean (occluded) | AEPE: 8.77 -> 7.91 | Improved |
| replace standard TV with unrolled cost | ARFlow · Sintel Clean (non-occluded) | AEPE: 1.27 -> 1.28 | No clear effect |
| replace standard TV with unrolled cost | SMURF · Sintel Final (non-occluded) | AEPE: 2.08 -> 2.11 | Hurt |
| use Charbonnier or Huber relaxation instead of standard TV | ARFlow · Sintel Clean | AEPE: 2.29 -> 2.34 (Charbonnier), 2.27 (Huber) | No clear effect |
| replace TV with unrolled cost in DIP denoising | DIP-TV · Image denoising (sigma=100) | PSNR (dB): 20.13 -> 20.22 | Improved |
| replace TV with unrolled cost in DIP denoising | DIP-TV · Image denoising (sigma=25) | PSNR (dB): 27.91 -> 27.94 | No clear effect |
| increase cost update steps T from 1 to 4 (ablation) | ARFlow finetuning · Sintel Final | AEPE: 3.25 -> 3.16 | Improved |
| increase update steps T from 2 to 4 (ablation) | ARFlow finetuning · Sintel Clean | AEPE: 2.35 -> 2.35 | No clear effect |
| replace standard TV with unrolled cost | ARFlow · training cost | memory: 4.326Gi -> 4.285Gi | Improved |
Closest papers by meaning
- What Matters in Unsupervised Optical Flow (ECCV 2020 · similarity 0.85)
- Unsupervised Deep Learning for Optical Flow Estimation (AAAI 2017 · similarity 0.84)
- Stacey: Promoting Stochastic Steepest Descent via Accelerated $\ell_p$-Smooth Nonconvex Optimization (ICML 2025 · similarity 0.84)
- Unrolled denoising networks provably learn to perform optimal Bayesian inference (NeurIPS 2024 · similarity 0.84)
- Penalizing Gradient Norm for Efficiently Improving Generalization in Deep Learning (ICML 2022 · similarity 0.84)
- A statistical perspective on algorithm unrolling models for inverse problems (JMLR 2025 · similarity 0.84)
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.
