Research map / 3D vision / Depth and stereo
LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow
Hongyu Wen, Erich Liang, Jia Deng
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 |
|---|---|---|---|
| fine-tune RAFT jointly on synthetic LayeredFlow data and Sintel (S+L) (ablation) | RAFT · LayeredFlow first layer, all points | EPE: 16.49 -> 15.63 | Improved |
| fine-tune RAFT on Sintel only (ablation) | RAFT · LayeredFlow first layer, all points | EPE: 16.49 -> 17.94 | No clear effect |
| fine-tune RAFT on synthetic data only (L) vs adding Sintel (ablation) | RAFT · LayeredFlow first layer, diffuse | EPE: 10.76 -> 14.69 (L); 6.95 (S+L) | Hurt |
| fine-tune RAFT on synthetic data for last-layer prediction | RAFT · LayeredFlow last layer, behind transparent | EPE: 8.48 -> 4.34 | Improved |
| fine-tune RAFT on synthetic data for last layer | RAFT · LayeredFlow last layer, all points | EPE: 8.28 -> 7.72 | Improved |
| fine-tune RAFT on synthetic data (L) for reflective first-layer points (ablation) | RAFT · LayeredFlow first layer, reflective | EPE: 8.51 -> 17.30 | Hurt |
| use multiple context encoders and layer heads (Multi-RAFT) instead of single-flow RAFT | RAFT · Multi-layer bad-3px, all points | bad-3px (%): 54.99 -> 48.98 | Improved |
| Multi-RAFT vs RAFT on diffuse surfaces | Multi-RAFT (L) · Multi-layer bad-tau, diffuse | bad-3px (%): 24.57 -> 28.75 | Hurt |
| add Sintel to Multi-RAFT fine-tuning (S+L) at higher layers (ablation) | Multi-RAFT · Multi-layer count-aware bad-tau, Layer 2 and 3 | bad-3px (%): 79.93 -> 74.93 (Layer 2); 88.50 -> 85.88 (Layer 3) | Improved |
Closest papers by meaning
- MPI-Flow: Learning Realistic Optical Flow with Multiplane Images (ICCV 2023 · similarity 0.88)
- On the real-world generalisability of Optical Flow models (ECCV 2026 · similarity 0.88)
- Seeing and Seeing Through the Glass: Real and Synthetic Data for Multi-Layer Depth Estimation (ICCV 2025 · similarity 0.87)
- Flow-Anything: Learning Real-World Optical Flow Estimation From Large-Scale Single-View Images (TPAMI 2025 · similarity 0.87)
- Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within Unified Neural Representations (CVPR 2025 · similarity 0.86)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · 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.
