AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Text-Guided Graph Neural Networks for Referring 3D Instance Segmentation

Pin-Hao Huang, Han-Hung Lee, Hwann-Tzong Chen, Tyng-Luh Liu

AAAI 2021 · Point clouds › Segmentation and detection · approach: Scene completion and instance segmentation · 118 citations (OpenAlex)

Read the paper

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 relation coordinate encoding to the GRU TGNN (ablation)TGNN · ScanRefer validationmIOU: 24.8 ± 0.4 -> 26.1 ± 0.2Improved
add instance-dependent language attention to the GRU TGNN (ablation)TGNN · ScanRefer validationmIOU: 25.6 ± 0.2 -> 26.1 ± 0.2Improved
add instance-dependent language attention to the BERT TGNN (ablation)TGNN · ScanRefer validationmIOU: 26.8 ± 0.3 -> 27.8 ± 0.2Improved
use relation encoding and language attention instead of simple feature concatenationTGNN with GRU · ScanRefer validationmIOU: 22.6 ± 0.3 -> 26.1 ± 0.2Improved
use BERT rather than GRU for language features in TGNNTGNN · ScanRefer validationmIOU: 26.1 ± 0.2 -> 27.8 ± 0.2Improved
use TGNN rather than ReferIt3D's modelreferring object identification · Nr3Daccuracy: 35.6% ± 0.7% -> 37.3% ± 0.3%Improved
use TGNN rather than ReferIt3D's modelreferring object identification · Sr3Daccuracy: 40.8% ± 0.2% -> 45.0% ± 0.2%Improved
use OursBERT rather than ScanRefer3D object localization · ScanRefer testAcc@0.25: 42.44 -> 41.00Hurt
use OursBERT rather than ScanRefer3D object localization · ScanRefer testAcc@0.5: 26.03 -> 32.80Improved

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