AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Sparse-to-Dense Feature Matching: Intra and Inter Domain Cross-Modal Learning in Domain Adaptation for 3D Semantic Segmentation

Duo Peng, Yinjie Lei, Wen Li, Pingping Zhang, Yulan Guo

ICCV 2021 · Point clouds › Segmentation and detection · approach: Scene completion and instance segmentation · 69 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 DsCML to baseline (ablation)xMUDA-style 2D/3D network · all five DA settingsmIoU: 2D +3.2 to +10, 3D +3.4 to +7.8, Avg +2.3 to +3.7Improved
add CMAL on top of DsCML (ablation)DsCML · all settingsmIoU: over 1.1 on 2D and over 2.1 on 3DImproved
add pseudo-labeling to DsCML+CMAL (ablation)DsCML + CMAL · USA->Singapore (nuScenes)3D mIoU: 55.6 -> 56.3Improved
use sparse-to-dense square patch (sCML) instead of one-to-one matching (CML) (ablation)DsCML · USA->Singapore (nuScenes)3D mIoU: 51.7 -> 52.5Improved
use deformable patch (DsCML) instead of square patch (sCML) (ablation)DsCML · A2D2->SemanticKITTI3D mIoU: 44.3 -> 45.1Improved
use max/min sparse-to-dense loss instead of patch-mean loss (ablation)DsCML · A2D2->SemanticKITTI2D mIoU: 37.6 -> 39.6Improved
add sparse-to-dense loss L_std (ablation)DsCML · USA->Singapore (nuScenes)2D mIoU: 53.4 -> 61.3Improved
use cross-modal alignment (S2D<->T3D, S3D<->T2D) instead of inter-modal alignment (S2D<->T2D) (ablation)CMAL · USA->Singapore (Lidarseg)3D mIoU: 54.9 -> 56.2Improved
use both alignment schemes together instead of cross-modal only (ablation)CMAL · USA->Singapore (Lidarseg)3D mIoU: 56.2 -> 55.3Hurt

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