Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• GeoGuide integrates hierarchical geometry-semantic consistency for open-vocabulary 3D segmentation by leveraging pretrained 3D models.method 95%
• The Uncertainty-based Superpoint Distillation (USD) module fuses geometric and semantic features to estimate per-point uncertainty, adaptively weighting 2D features within superpoints.method 95%
• The Instance-level Mask Reconstruction module uses geometric priors to enforce semantic consistency within instances by reconstructing complete instance masks.method 95%
• The Inter-Instance Relation Consistency module aligns geometric and semantic similarity matrices to calibrate cross-instance consistency for same-category objects, mitigating viewpoint-induced semantic drift.method 95%
• GeoGuide demonstrates superior performance on ScanNet v2, Matterport3D, and nuScenes datasets.result 95%
• GeoGuide achieves results comparable to or exceeding early fully-supervised approaches on ScanNet v2 while requiring no 3D annotations.result 90%
• Existing open-vocabulary 3D semantic segmentation methods predominantly rely on distilling knowledge from 2D open-vocabulary models, which restricts intrinsic 3D geometric learning and inherits errors from 2D predictions.limitation 95%
• GeoGuide utilizes a pure 3D model and does not use time-consuming self-distillation strategies.technical 90%
• GeoGuide integrates hierarchical geometry-semantic consistency for open-vocabulary 3D segmentation by leveraging pretrained 3D models.method 95%
• The Uncertainty-based Superpoint Distillation (USD) module fuses geometric and semantic features to estimate per-point uncertainty and adaptively weight 2D features within superpoints.method 95%
• The Instance-level Mask Reconstruction module uses geometric priors to enforce semantic consistency within instances by reconstructing complete instance masks.method 95%
• The Inter-Instance Relation Consistency module aligns geometric and semantic similarity matrices to calibrate cross-instance consistency for same-category objects, mitigating viewpoint-induced semantic drift.method 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card