Research map / Continual learning / Class-incremental
Continual Learning and Unknown Object Discovery in 3D Scenes via Self-Distillation
Mohamed El Amine Boudjoghra, Jean Lahoud, Salman Khan, Hisham Cholakkal, Rao M Anwer, Fahad Shahbaz Khan
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 exemplar replay and student auto-labeling with exemplar-free self-distillation from the previous-task model | 3D-OWIS · ScanNet200 Split C Task T2 | U-Recall: 15.80 -> 41.31 | Improved |
| replace exemplar replay with self-distillation | 3D-OWIS · ScanNet200 Split C Task T2 | mAP previously known: 13.90 -> 26.02 | Improved |
| self-distillation versus 3D-OWIS on wilderness impact | OpenDistill3D · ScanNet200 Split A Task T2 | WI: 0.007 -> 0.352 | Hurt |
| add Unknown-Known Separation (UKS) to naive self-distillation (ablation) | OpenDistill3D · ScanNet200 Split C Task T2 | mAP all classes: 21.83 -> 23.30 | Improved |
| add Unknown-Known Separation (UKS) (ablation) | OpenDistill3D · ScanNet200 Split C Task T2 | U-Recall: 36.57 -> 39.53 | Improved |
| add objectness weight function WI to cross-entropy loss (ablation) | OpenDistill3D · ScanNet200 Split C Task T2 | mAP all classes: 23.30 -> 24.24 | Improved |
| use objectness score itself as weight instead of WI (ablation) | OpenDistill3D · ScanNet200 Split C Task T1 | mAP all classes: 24.90 -> 22.08 | Hurt |
| use objectness score itself as weight instead of WI (ablation) | OpenDistill3D · ScanNet200 Split A Task T1 | mAP all classes: 38.81 -> 39.09 | Mixed |
| add unknown-object self-distillation to closed-set Mask3D training | Mask3D · ScanNet200 val (w/o DBSCAN) | mAP: 25.73 -> 27.37 | Improved |
| use OpenDistill3D as proposal network for OpenMask3D | OpenMask3D · Replica | mAP50: 15.90 -> 18.40 | Improved |
Closest papers by meaning
- Continual Segmentation with Disentangled Objectness Learning and Class Recognition (CVPR 2024 · similarity 0.87)
- MIRACLE 3D: Memory-efficient Integrated Robust Approach for Continual Learning on 3D Point Clouds via Shape Model Construction (ICLR 2025 · similarity 0.87)
- C-NAV: Towards Self-Evolving Continual Object Navigation in Open World (NeurIPS 2025 · similarity 0.86)
- ProtoDepth: Unsupervised Continual Depth Completion with Prototypes (CVPR 2025 · similarity 0.86)
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.86)
- Online Segment Any 3D Thing as Instance Tracking (NeurIPS 2025 · 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.
