Research map / Point clouds / Segmentation and detection
Scene completion and instance segmentation
Model object relations and open-vocabulary contrast for 3D instance segmentation and scene completion.
44 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Segmentation and detection, keyword cluster “semantic segmentation · cloud semantic · instance”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
13 papers in 2023–24 and 19 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 13.0% to 20.9%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · 3D Semantic Parsing of Large-Scale Indoor Spaces (CVPR 2016 · 1,979 citations)
- 2017 · Unsupervised Feature Learning for 3D Scene Reconstruction with Occupancy Maps (AAAI 2017 · 7 citations)
- 2018 · 3D Semantic Segmentation With Submanifold Sparse Convolutional Networks (CVPR 2018 · 1,861 citations)
- 2021 · Exploring Data-Efficient 3D Scene Understanding With Contrastive Scene Contexts (CVPR 2021 · 277 citations)
- 2022 · SoftGroup for 3D Instance Segmentation on Point Clouds (CVPR 2022 · 278 citations)
- 2023 · CLIP2Scene: Towards Label-Efficient 3D Scene Understanding by CLIP (CVPR 2023 · 153 citations)
- 2024 · OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · 66 citations)
- 2025 · BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025 · 13 citations)
- 2026 · Point Cloud Semantic Scene Completion with Prototype-Guided Transformer (AAAI 2026 · 4 citations)
Earliest papers here
- 3D Semantic Parsing of Large-Scale Indoor Spaces (CVPR 2016 · 1,979 citations)
- Semantic Instance Annotation of Street Scenes by 3D to 2D Label Transfer (CVPR 2016 · 160 citations)
- Unsupervised Feature Learning for 3D Scene Reconstruction with Occupancy Maps (AAAI 2017 · 7 citations)
What has been tried
31 papers here report 372 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 7%, hurt 14%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add adaptive aggregation rather than single-scale processing | ScanNet v2 val · mIoU | 75.0 -> 76.1 | Helped | OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · 66 citations) |
| Use Segment Grouping rather than matched-point positive pairs | ScanNet semantic segmentation · FT mIoU (%) | 74.6 -> 75.7 | Helped | GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding (CVPR 2024 · 23 citations) |
| Add the boundary-semantic block to the OctFormer baseline | ScanNet200 validation · mIoU | 32.7 -> 36.4 | Helped | BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025 · 13 citations) |
| add ASAM | ScanNetV2 validation · mAP | 59.8 -> 60.1 | Helped | Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · 6 citations) |
| Remove prototype-selection post-processing | Urban · Cham. | 0.29 -> 0.28 | Helped | Learnable Earth Parser: Discovering 3D Prototypes in Aerial Scans (CVPR 2024 · 6 citations) |
| replace random click sampling with iterative training | ScanNet40 · mean IoU@mean 10 | 82.0 -> 84.4 | Helped | AGILE3D: Attention Guided Interactive Multi-object 3D Segmentation (ICLR 2024 · 5 citations) |
| use cross-view augmentation and point-to-entity loss | ScanNet OV-SemSeg · f-mIoU | 63.6 -> 64.6 | Helped | Masked Point-Entity Contrast for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · 4 citations) |
| Add PWCA to the module-ablation baseline | ScanRefer · mIoU | 32.66 -> 33.44 | Helped | LESS: Label-Efficient and Single-Stage Referring 3D Segmentation (NeurIPS 2024 · 3 citations) |
| Add IKA to the kernel-based method | ScanNetV2 Val and S3DIS Area 5 · mAP | 56.8 -> 62.7; 56.3 -> 60.4 | Helped | Insightful Instance Features for 3D Instance Segmentation (CVPR 2025 · 2 citations) |
| adding the 3D completion subnetwork (without CVP) to BEV+segmentation branches | SemanticKITTI validation · IoU | +0.8 IoU (58.3 to 59.1) | Helped | Voxel Proposal Network via Multi-Frame Knowledge Distillation for Semantic Scene Completion (NeurIPS 2024 · 2 citations) |
| Add TSD while retaining GSS and SMS | SN2 synthetic benchmark · AUROC | 93.7 -> 94.8 | Helped | SASep: Saliency-Aware Structured Separation of Geometry and Feature for Open Set Learning on Point Clouds (CVPR 2025 · 1 citation) |
| Add scene-specific and object-specific latent variables | S3DIS · mIoU@10 | 88.7 -> 90.9 | Helped | Probabilistic Interactive 3D Segmentation with Hierarchical Neural Processes (ICML 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ScanNet200 · 4 of 7 papers
- ScanNetV2 · 4 of 7 papers
- S3DIS · 3 of 7 papers
- ScanNet · 2 of 7 papers
Baselines
- ISBNet · 2 of 7 papers
- MAFT · 2 of 7 papers
- Mask3D · 2 of 7 papers
- Spherical Mask · 2 of 7 papers
- SoftGroup · 2 of 7 papers
- SPFormer · 2 of 7 papers
- PointGroup · 2 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 32 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 4 papers
- Huazhong University of Science and Technology · 2 papers
- The University of Hong Kong · 2 papers
- HKU · 2 papers
- Meta · 2 papers
- South China University of Technology · 2 papers
- Shenzhen University · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Point cloud semantic segmentation (108 papers)
- LiDAR segmentation for driving (47 papers)
Elsewhere in the topic
- LiDAR voxel and pillar detectors (56 papers)
- Single-stage point-voxel detection (45 papers)
- Multimodal and open-vocabulary 3D detection (24 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).