Research map / 3D vision / Gaussian splatting
Easy3D: A Simple Yet Effective Method for 3D Interactive Segmentation
Andrea Simonelli, Norman Müller, Peter Kontschieder
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 |
|---|---|---|---|
| use implicit click fusion instead of explicit fusion (ablation) | Easy3D · KITTI-360 (trained on ScanNet40) | IoU@10: 48.2 -> 83.6 | Improved |
| use implicit click fusion instead of explicit fusion (ablation) | Easy3D · KITTI-360 (trained on ScanNet40) | IoU@1: 34.5 -> 46.3 | Improved |
| use implicit click fusion instead of explicit fusion (ablation) | Easy3D · S3DIS (trained on ScanNet40) | IoU@1: 58.2 -> 65.7 | Improved |
| use implicit click fusion instead of explicit fusion (ablation) | Easy3D · ScanNet40 (trained on ScanNet40) | IoU@10: 83.6 -> 81.7 | Hurt |
| add learned negative embedding to implicit fusion (ablation) | Easy3D · KITTI-360 (trained on ScanNet40) | IoU@1: 44.9 -> 46.3 | Improved |
| add learned negative embedding to implicit fusion (ablation) | Easy3D · ScanNet40 (trained on ScanNet40) | IoU@1: 66.4 -> 68.2 | Improved |
| add learned negative embedding to explicit fusion (ablation) | Easy3D · S3DIS (trained on ScanNet40) | IoU@1: 53.8 -> 58.2 | Improved |
| add learned negative embedding to explicit fusion (ablation) | Easy3D · KITTI-360 (trained on ScanNet40) | IoU@1: 31.0 -> 34.5 | Improved |
| train on Gaussian Splatting scenes (GS->GS) rather than mesh (Mesh->GS) | Easy3D · GS-ScanNet40 | IoU@1: 44.9 -> 63.9 | Improved |
| Easy3D versus AGILE3D on single-click unseen+seen classes | Easy3D · ScanNet20->ScanNet40 | mAP: 24.8 -> 39.2 | Improved |
| Easy3D versus AGILE3D with many clicks on ScanNet++ | Easy3D · ScanNet++ (trained on ScanNet40) | IoU@10: 74.2 -> 71.3 | Mixed |
| voxel-based encoder (Easy3D) versus point-group ViT encoder (Point-SAM) | Easy3D · S3DIS (trained on ScanNet40) | IoU@1: 38.8 -> 65.7 | Improved |
Later papers that compared with this method
1 later accepted paper reports Easy3D as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- NegROI: Click-Centric Uncertainty-Guided Refinement with Scene-Conditioned Negative Prompts for Robust Interactive 3D Segmentation (ECCV 2026 · better on 7 of 8 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
- AGILE3D: Attention Guided Interactive Multi-object 3D Segmentation (ICLR 2024 · similarity 0.91)
- Probabilistic Interactive 3D Segmentation with Hierarchical Neural Processes (ICML 2025 · similarity 0.88)
- COS3D: Collaborative Open-Vocabulary 3D Segmentation (NeurIPS 2025 · similarity 0.87)
- ReferSplat: Referring Segmentation in 3D Gaussian Splatting (ICML 2025 · similarity 0.86)
- GaussianCut: Interactive segmentation via graph cut for 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.86)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.86)
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.
