AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Separable Flow: Learning Motion Cost Volumes for Optical Flow Estimation

Feihu Zhang, Oliver J. Woodford, Victor Adrian Prisacariu, Philip H.S. Torr

ICCV 2021 · Architectures and efficiency › CNN design · approach: Dense matching and estimation CNNs · 113 citations (OpenAlex)

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 changedWhereMeasuredEffect
add the separable cost volume moduleRAFT · KITTI-15 train (trained on C+T)Fl-all (%): 17.4 -> 15.9Improved
add the separable cost volume modulePWC-Net · KITTI-15 train (trained on C+T)Fl-all (%): 33.7 -> 23.1Improved
add the separable cost volume moduleRAFT · Sintel train clean (trained on C+T)EPE: 1.43 -> 1.30Improved
add extra Virtual KITTI synthetic training dataSeparable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 15.9 -> 7.74Improved
add attention channels to mean and max separation (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.7 -> 15.9Improved
add attention channels to max-only separation (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.5 -> 15.9Improved
use 4x SGA layers instead of 2x 3D conv aggregation (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.8 -> 15.9Improved
use 4x SGA layers instead of 8x 3D conv (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.4 -> 15.9Improved
use separate aggregation weights for u and v instead of shared (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.3 -> 15.9Improved
use UNet aggregation block instead of ResNet (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.1 -> 15.9Improved
add motion regression initialization (ablation)Separable Flow · KITTI-15 train (trained on C+T)Fl-all (%): 16.8 -> 15.9Improved
add motion regression initialization (ablation)Separable Flow · Sintel train clean (trained on C+T)EPE: 1.37 -> 1.30Improved

Later papers that compared with this method

3 later accepted papers report SeparableFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy