AI Paper Atlas

Research map / 3D vision / 3D generation

OVSeg3R: Learn Open-vocabulary Instance Segmentation from 2D via 3D Reconstruction

Hongyang Li, Jinyuan Qu, Lei Zhang

ICLR 2026 · 3D vision › 3D generation · approach: Learning 3D reconstruction from images · 3 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 View-wise Instance Partition (ablation)OVSeg3R · ScanNet200 standard settingmAP: 18.4 -> 23.9Improved
add View-wise Instance Partition (ablation)OVSeg3R · ScanNet200 standard settingmAP novel: 16.2 -> 23.0Improved
use image-based superpoints (IBSp) instead of geometric-only superpoints (ablation)OVSeg3R · ScanNet200 standard settingmAP: 23.6 -> 23.9Improved
increase reconstructed data volume from 0% to 100% (ablation)OVSeg3R · ScanNet200 standard settingmAP: 5.0 -> 23.9Improved
increase reconstructed data volume from 0% to 100% (ablation)OVSeg3R · ScanNet200 standard settingmAP novel: 4e-4 -> 23.0Improved
add ScanNet3R-VGGT data alongside MASt3R-SLAM data (ablation)OVSeg3R · ScanNet200 standard settingmAP: 23.9 -> 24.6Improved
filter out failed reconstructions (ablation)OVSeg3R · ScanNet200 standard settingmAP: 23.0 -> 23.9Improved
use DINO-X universal mode features instead of 20-class prompts (ablation)OVSeg3R · ScanNet200 standard settingmAP novel: 14.0 -> 23.0Improved
use DINO-X universal mode features instead of 20-class prompts (ablation)OVSeg3R · ScanNet200 standard settingmAP base: 23.9 -> 26.2Improved
train SegDINO3D-VL with OVSeg3R instead of ScanNet200 onlySegDINO3D-VL · ScanNet200 val (open setting)mAP: 38.4 -> 40.7Improved
train SegDINO3D-VL with OVSeg3R instead of ScanNet200 onlySegDINO3D-VL · ScanNet200 val (Tail classes)mAP: 34.0 -> 42.7Improved
replace classification head with text-similarity head, trained on ScanNet200 onlySegDINO3D · ScanNet200 valmAP: 39.8 -> 38.4Hurt

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