AI Paper Atlas

Research map / 3D vision / Depth and stereo

Flow3r: Factored Flow Prediction for Scalable Visual Geometry Learning

Zhongxiao Cong, Qitao Zhao, Minsik Jeon, Shubham Tulsiani

CVPR 2026 · 3D vision › Depth and stereo · approach: Learned optical and scene flow · 7 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
use factored flow prediction instead of VGGT-style tracking head (ablation)small VGGT-like · ScanNet++ staticRTA@30: 0.7021 -> 0.7366Improved
use factored flow instead of tracking head (ablation)small VGGT-like · OmniWorld dynamicRRA@30: 68.56 -> 76.26Improved
supervise projection-based flow from decoded cameras and depth (ablation)small VGGT-like · ScanNet++ staticRTA@30: 0.6929 -> 0.4572Hurt
tracking-style flow supervision on unlabeled data (ablation)small VGGT-like · ScanNet++ staticRTA@30: 0.6929 -> 0.7021No clear effect
add unlabeled videos with factored flow supervision (ablation)small VGGT-like · OmniWorld dynamicRRA@30: 66.01 -> 76.26Improved
scale unlabeled SpatialVID from 3K to 20K (ablation)small VGGT-like · OmniWorld dynamicRRA@30: 76.26 -> 81.12Improved
use 20K unlabeled videos instead of 3K extra labeled sequences (4K labeled) (ablation)small VGGT-like · OmniWorld dynamicRRA@30: 78.68 -> 81.12Improved
add labeled multi-task flow head to pi3 (ablation)pi3 · 7-scenesRTA: 87.27 -> 87.99No clear effect
add unlabeled videos via flow supervision to pi3 (ablation)pi3 · 7-scenesRTA: 87.99 -> 91.66Improved
add unlabeled videos via flow supervision to VGGT (ablation)VGGT · 7-scenesRTA: 89.17 -> 90.67Improved
train Flow3r (pi3 + unlabeled data) vs pi3pi3 · Sintel (dynamic)RPE trans: 0.066 -> 0.058Improved
train Flow3r vs pi3 on static datasetspi3 · ScanNetAUC@30: 69.39 -> 71.00 (VGGT 93.71 RTA vs 92.89)Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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