AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Spatial Matters: Position-Guided 3D Referring Expression Segmentation

Yabing Wang, Zhuotao Tian, Le Wang, Zheng Qin, Sanping Zhou

CVPR 2026 · Point clouds › Segmentation and detection · approach: Scene completion and instance segmentation

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
build point proxies (local context plus spatial interaction) before query selection instead of using FPS key points directly (ablation)Position3D · ScanRefermIoU: 48.2 -> 51.0Improved
add spatial relation modeling in query generation (ablation)Position3D · ScanRefermIoU: 49.2 -> 51.0Improved
add spatial relation modeling in the decoder (ablation)Position3D · ScanRefermIoU: 49.0 -> 51.0Improved
use progressively decreasing super-points {128,64,32,16} instead of all super-points (ablation)position-guided deformable attention · ScanRefermIoU: 49.4 -> 51.0Improved
use constant 16 super-points per layer (ablation)position-guided deformable attention · ScanRefermIoU: 51.0 -> 48.2Hurt
shrink super-points faster {128,64,16,8} (ablation)position-guided deformable attention · ScanRefermIoU: 51.0 -> 50.3Hurt
use 8 nearest neighbors in local context aggregation (ablation)Position3D · ScanRefermIoU: 51.0 -> 49.0Hurt
use 64 nearest neighbors in local context aggregation (ablation)Position3D · ScanRefermIoU: 51.0 -> 49.5Hurt
Position3D vs IPDN on unique subsetPosition3D · ScanRefer (Unique subset)mIoU: 78.6 -> 77.6Hurt
Position3D vs IPDN on multiple subsetPosition3D · ScanRefer (Multiple subset)mIoU: 43.6 -> 44.7Improved
Position3D vs MDIN on zero-target with distractorsPosition3D · Multi3DRefer (zero target with distractors)Acc@0.25: 47.9 -> 45.5Hurt

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