Research map / Architectures and efficiency / RNNs and normalization
Volumetric Correspondence Networks for Optical Flow
Gengshan Yang, Deva Ramanan
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 true 4D convolution on cost volume instead of 2D DenseNet matching (ablation) | single-stage matching module · KITTI-15 train 0-32px | EPE: 2.64 -> 2.30 | Improved |
| separate 4D kernels into spatial and WTA 2D kernels (ablation) | Full-4D · KITTI-15 train 0-32px | EPE: 2.30 -> 2.31 | No clear effect |
| use volumetric U-Net instead of plain 4D conv stack (ablation) | Ours-UNet · KITTI-15 train 0-32px | EPE: -0.02 EPE for plain, +20.9 GFlops | No clear effect |
| add multi-channel cost volume (ablation) | Ours-UNet · KITTI-15 train 0-32px | EPE: +0.32 when removed | Improved |
| use truncated soft-argmin instead of soft-argmin (ablation) | Ours-UNet · KITTI-15 train 0-32px | EPE: +0.10 when replaced | Improved |
| use truncated soft-argmin instead of direct regression (ablation) | Ours-UNet · KITTI-15 train 0-32px | EPE: +0.58 when replaced | Improved |
| add out-of-range detection module (ablation) | Ours-UNet · KITTI-15 train 0-32px | EPE: +0.07 when removed | Improved |
| VCN vs PWC-Net when extending search range beyond training range | flow network · KITTI-15 train 0-64px | EPE: 9.44 -> 3.27 | Improved |
| VCN vs best prior fine-tuned methods | optical flow · Sintel test clean | EPE: 3.45 -> 2.81 | Improved |
| VCN vs PWC-Net on stereo without stereo training | flow network · KITTI-15 stereo train | D1-all: 23.30 -> 8.73 | Improved |
| use 2D search window for stereo with imperfect rectification | stereo matching · Middlebury-14 | avgerr: 9.92 (PWC-Net) -> 8.79 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PWC-Net — PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume (CVPR 2018) · better on 3 of 3 shared results
- LiteFlowNet2 — LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation (CVPR 2018) · better on 1 of 1 shared result
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
- CFNet: Cascade and Fused Cost Volume for Robust Stereo Matching (CVPR 2021 · similarity 0.87)
- LiteFlowNet3: Resolving Correspondence Ambiguity for More Accurate Optical Flow Estimation (ECCV 2020 · similarity 0.87)
- SMD-Nets: Stereo Mixture Density Networks (CVPR 2021 · similarity 0.87)
- Displacement-Invariant Matching Cost Learning for Accurate Optical Flow Estimation (NeurIPS 2020 · similarity 0.87)
- Learning Stereo from Single Images (ECCV 2020 · similarity 0.86)
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.85)
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.
