Research map / Architectures and efficiency / RNNs and normalization
Deep Equilibrium Optical Flow Estimation
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 |
|---|---|---|---|
| Replace recurrent updates with a DEQ fixed-point formulation | RAFT · Sintel (test), Clean | AEPE: 1.94 -> 1.82 | Helped |
| Replace recurrent updates with a DEQ fixed-point formulation | RAFT · Sintel (test), Final | AEPE: 3.18 -> 3.23 | Hurt |
| Replace recurrent updates with a DEQ fixed-point formulation | RAFT · KITTI-15 (test) | F1-all: 5.10 -> 4.91 | Helped |
| Increase hidden dimensions to the DEQ-RAFT-H model | DEQ-RAFT · KITTI-15 (train), C+T | F1-all: 17.4 -> 14.9 | Helped |
| Add 3-step phantom gradient (ablation) | DEQ-RAFT-H · KITTI-15 (train), C+T | F1-all: 14.9 -> 13.5 | Helped |
| Add sparse fixed-point correction loss (ablation) | DEQ flow · FlyingChairs validation | AEPE: about 9% | Helped |
| Add sparse fixed-point correction loss (ablation) | DEQ flow · FlyingChairs validation | fixed-point residual error: over 60% | Helped |
| Use one-step inexact gradients with fixed-point correction instead of IFT training (ablation) | DEQ flow · FlyingChairs training | training time: over 45% | Helped |
| Use Jacobian regularization instead of fixed-point correction (ablation) | DEQ flow · FlyingChairs validation | AEPE | Hurt |
| Increase fixed-point correction frequency beyond sparse corrections (ablation) | DEQ flow · FlyingChairs validation | AEPE | No clear effect |
| Reuse the previous frame's fixed point as the next solver initialization | DEQ flow · Sintel video sequences | inference speed: about 1.6 × | Helped |
Closest papers by meaning
- FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · similarity 0.88)
- DeltaDEQ: Exploiting Heterogeneous Convergence for Accelerating Deep Equilibrium Iterations (NeurIPS 2024 · similarity 0.88)
- Context-Aware Iteration Policy Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.87)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.87)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.87)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 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.