Research map / Point clouds / Segmentation and detection
SAQN: Semantic-based Adaptive Query Network for 3D Referring Expression Segmentation
Jiale Huang, Shangfei Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace instance-based queries with semantic-class queries (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | zero-target Acc@0.5: 37.1 -> 54.9 | Improved |
| replace instance-based queries with semantic-class queries (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | multi-target Acc@0.5: 51.2 -> 49.6 | Hurt |
| replace instance-based queries with semantic-class queries (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 50.3 -> 52.1 | Improved |
| add Adaptive Query Fusion on top of semantic queries (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | multi-target Acc@0.5: 49.6 -> 51.1 | Improved |
| add Adaptive Queries Intersection Loss on top of SQ + AQF (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 52.5 -> 53.1 | Improved |
| add AQF and AQIL to instance-based queries (without semantic queries) (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | zero-target Acc@0.5: 37.1 -> 36.9 | No clear effect |
| use only adaptive queries (31) with no semantic prior (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 53.1 -> 50.4 | Hurt |
| vary adaptive query number k2 from 10 to 15 (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 53.1 -> 51.9 | Hurt |
| vary adaptive query number k2 from 10 to 5 (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 53.1 -> 52.7 | Mixed |
| raise loss weight lambda_a from 1 to 10 (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 53.1 -> 49.9 | Hurt |
| lower loss weight lambda_a from 1 to 0.1 (ablation) | SAQN · Multi3DRefer (3D-GRES), distractors | mIoU: 53.1 -> 52.3 | Hurt |
| semantic-class queries versus instance-based queries (IPDN), strict threshold | SAQN · ScanRefer (3D-RES) overall | Acc@0.5: 54.9 -> 53.4 | Hurt |
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.
- IPDN — IPDN: Image-enhanced Prompt Decoding Network for 3D Referring Expression Segmentation (AAAI 2025) · better on 4 of 7 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
- 3D-STMN: Dependency-Driven Superpoint-Text Matching Network for End-to-End 3D Referring Expression Segmentation (AAAI 2024 · similarity 0.90)
- LESS: Label-Efficient and Single-Stage Referring 3D Segmentation (NeurIPS 2024 · similarity 0.90)
- IPDN: Image-enhanced Prompt Decoding Network for 3D Referring Expression Segmentation (AAAI 2025 · similarity 0.90)
- RG-SAN: Rule-Guided Spatial Awareness Network for End-to-End 3D Referring Expression Segmentation (NeurIPS 2024 · similarity 0.89)
- X-RefSeg3D: Enhancing Referring 3D Instance Segmentation via Structured Cross-Modal Graph Neural Networks (AAAI 2024 · similarity 0.87)
- Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · similarity 0.87)
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.
