Research map / Detection and segmentation / 3D detection and driving
EvObj: Learning Evolving Object-centric Representations for 3D Instance Segmentation without Scene Supervision
Jiahao Chen, Zihui Zhang, Yafei Yang, Jinxi Li, Shenxing Wei, Zhixuan Sun, Bo Yang
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 |
|---|---|---|---|
| add object candidate discerning module (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 45.0 -> 55.0 | Improved |
| add evolving optimization of discerning module (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 52.2 -> 55.0 | Improved |
| add synthetic pretraining of discerning module (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 37.4 -> 55.0 | Improved |
| add object completion module (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 33.8 -> 55.0 | Improved |
| evolving interval T=50 instead of 100 (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 52.6 | Hurt |
| evolving interval T=200 instead of 100 (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 53.4 | Hurt |
| use PoinTr instead of AdaPoinTr for completion (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 53.1 | No clear effect |
| use SnowflakeNet instead of AdaPoinTr for completion (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 54.1 | No clear effect |
| use PointNet++ instead of SparseUNet for discerning module (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 54.5 | No clear effect |
| Chamfer threshold 0.08 instead of 0.10 (ablation) | EvObj-VAE · ScanNet val (chair) | AP: 55.0 -> 48.0 | Hurt |
| apply discerning module as plug-and-play refinement | UnScene3D · ScanNet val (chair) | AP: 37.2 -> 41.1 | Improved |
| apply discerning module as plug-and-play refinement | Part2Object · ScanNet val (chair) | AP: 34.4 -> 39.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UnScene3D — UnScene3D: Unsupervised 3D Instance Segmentation for Indoor Scenes (CVPR 2024) · better on 3 of 3 shared results
- EFEM — EFEM: Equivariant Neural Field Expectation Maximization for 3D Object Segmentation Without Scene Supervision (CVPR 2023) · better on 2 of 2 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
- FoundObj: Self-supervised Foundation Models as Rewards for Label-free 3D Object Segmentation (ICML 2026 · similarity 0.88)
- GrabS: Generative Embodied Agent for 3D Object Segmentation without Scene Supervision (ICLR 2025 · similarity 0.87)
- Unsupervised 3D Object Segmentation of Point Clouds by Geometry Consistency (TPAMI 2024 · similarity 0.86)
- Part2Object: Hierarchical Unsupervised 3D Instance Segmentation (ECCV 2024 · similarity 0.86)
- UnScene3D: Unsupervised 3D Instance Segmentation for Indoor Scenes (CVPR 2024 · similarity 0.85)
- Living Scenes: Multi-object Relocalization and Reconstruction in Changing 3D Environments (CVPR 2024 · similarity 0.84)
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.
