AI Paper Atlas

Research map / 3D vision / Depth and stereo

Parameterized Cost Volume for Stereo Matching

Jiaxi Zeng, Chengtang Yao, Lidong Yu, Yuwei Wu, Yunde Jia

ICCV 2023 · 3D vision › Depth and stereo · approach: Learned stereo with specialized sensors · 39 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
replace single Gaussian fixed variance with multi-Gaussian adaptive variance (parameterized cost volume) (ablation)RAFT-Stereo · SceneFlow, 4th iterationEPE (px): 0.79 -> 0.72Improved
add adaptive variance to a single Gaussian (ablation)RAFT-Stereo (SGFV) · SceneFlow, 4th iterationEPE (px): 0.79 -> 0.82Hurt
use multi-Gaussian with fixed variance instead of single Gaussian (ablation)RAFT-Stereo (SGFV) · SceneFlow, 4th iteration1px-error rate (%): 8.96 -> 8.93No clear effect
make multi-Gaussian variance adaptive (ablation)MGFV · SceneFlow, 4th iteration1px-error rate (%): 8.93 -> 8.16Improved
add uncertainty-aware refinement module (ablation)PCV · SceneFlow, 4th iteration1px-error rate (%): 8.16 -> 7.98Improved
add uncertainty-aware refinement module (ablation)PCV · SceneFlow, 4th iterationEPE (px): 0.72 -> 0.71Improved
use PCV instead of dynamic cost volume at same 4 iterationsRAFT-Stereo (T=4) · SceneFlowEPE (px): 0.79 -> 0.71Improved
use PCV with 4 iterations instead of RAFT-Stereo with 32 iterationsRAFT-Stereo · SceneFlowinference time (ms): 205 -> 41Improved
use PCV with 4 iterations instead of RAFT-Stereo with 4 iterationsRAFT-Stereo (T=4) · SceneFlowFLOPS (G): 678.45 -> 798.23Hurt
use PCVNet instead of RAFT-Stereo on BoosterRAFT-Stereo · BoosterBad 2 (%): 38.66 -> 27.41Improved
use PCVNet instead of CREStereo on KITTI 2015 for speedCREStereo · KITTI 2015D1-all (%): 1.69 -> 1.93; 410 ms -> 56 msMixed

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