SCOPE: Scene-Contextualized Incremental Few-Shot 3D Segmentation

Vishal Thengane; Zhaochong An; Tianjin Huang; Son Lam Phung; Abdesselam Bouzerdoum; Lu Yin; Na Zhao; Xiatian Zhu

cs.CVMar 6, 2026

SCOPE: Scene-Contextualized Incremental Few-Shot 3D Segmentation

Authors: Vishal Thengane, Zhaochong An, Tianjin Huang, Son Lam Phung, Abdesselam Bouzerdoum, Lu Yin, Na Zhao, Xiatian Zhu

Organizations: University of Surrey, UK · University of Wollongong, Australia · University of Copenhagen, Denmark · University of Exeter, UK ·

Abstract

Incremental Few-Shot (IFS) segmentation aims to learn new categories over time from only a few annotations. Although widely studied in 2D, it remains underexplored for 3D point clouds. Existing methods suffer from catastrophic forgetting or fail to learn discriminative prototypes under sparse supervision, and often overlook a key cue: novel categories frequently appear as unlabelled background in base-training scenes. We introduce SCOPE (Scene-COntextualised Prototype Enrichment), a plug-and-play background-guided prototype enrichment framework that integrates with any prototype-based 3D segmentation method. After base training, a class-agnostic segmentation model extracts high-confidence pseudo-instances from background regions to build a prototype pool. When novel classes arrive with few labelled samples, relevant background prototypes are retrieved and fused with few-shot prototypes to form enriched representations without retraining the backbone or adding parameters. Experiments on ScanNet and S3DIS show that SCOPE achieves SOTA performance, improving novel-class IoU by up to 6.98% and 3.61%, and mean IoU by 2.25% and 1.70%, respectively, while maintaining low forgetting. Code is available https://github.com/Surrey-UP-Lab/SCOPE.

Figures & tables

Figure 1: Compared with GW [ 46 ] , SCOPE improves performance on novel classes by enriching prototypes with background context, without additional computational overhead.

PCS Paradigms	Future Knowledge	Few-Shot	Base + Novel	Multi-Step
Fully Supervised	–	✗	–	–
Few-Shot	✗	✓	✗	✗
Generalized Few-Shot	✗	✓	✓	✗
Class-Incremental	✓	✗	✓	✓
Incremental Few-Shot	✓	✓	✓	✓

Table 1: Comparison of PCS paradigms across future knowledge, few-shot, base-novel generalisation, and multi-step learning.

Figure 2: Overview of SCOPE . The framework comprises three stages: (1) Base Training , where the encoder $\Phi$ is trained on labelled base data; (2) Scene Contextualisation , which extracts background regions to build an Instance Prototype Bank (IPB); and (3) Incremental Class Registration , where retrieved prototypes are fused with few-shot prototypes via attention to yield refined novel-class representations.

Methods	Venue	$K{=}5$						$K{=}1$
Methods	Venue	mIoU	mIoU-B	mIoU-N	HM	mIOU-I	FPP	mIoU	mIoU-B	mIoU-N	HM	mIOU-I	FPP
JT (Oracle)	–	45.34	48.68	36.97	42.03	–	–	45.34	48.68	36.97	42.03	–	–
Finetuning	–	0.38	0.48	0.15	0.23	12.28	34.25	0.42	0.49	0.26	0.34	10.93	40.67
LwF [ 24 ]	TPAMI’17	4.12	5.77	0.27	0.51	14.75	35.39	0.37	0.33	0.47	0.39	12.06	40.84
CLIMB-3D [ 39 ]	BMVC’25	4.23	5.92	0.02	0.04	15.98	35.25	0.02	0.00	0.06	0.00	10.36	41.16
EWC [ 21 ]	PNAS’17	3.87	5.41	0.03	0.06	17.66	35.76	2.33	4.42	0.02	0.05	16.57	35.22

Table 2: Comparison of baseline methods and SCOPE on the ScanNet dataset under the IFS-PCS setting with

K{=}5

and

K{=}1

. We report mIoU, mIoU-B, mIoU-N, HM, mIoU-I, and FPP. All metrics are higher is better except FPP. Best results are highlighted in bold .

Methods	Venue	$K{=}5$						$K{=}1$
Methods	Venue	mIoU	mIoU-B	mIoU-N	HM	mIOU-I	FPP	mIoU	mIoU-B	mIoU-N	HM	mIOU-I	FPP
JT (Oracle)	–	73.62	81.57	64.34	71.94	–	–	73.62	81.57	64.34	71.94	–	–
Finetuning	–	2.25	3.26	1.07	1.61	24.57	65.55	0.52	0.06	1.06	0.11	22.88	75.58
LwF [ 24 ]	TPAMI’17	12.66	20.29	3.76	6.35	37.55	55.36	7.89	12.61	2.39	4.02	28.44	63.04
CLIMB-3D [ 39 ]	BMVC’25	17.71	31.37	1.76	3.32	41.72	44.27	7.24	12.77	0.79	1.49	30.98	68.40
EWC [ 21 ]	PNAS’17	19.48	35.56	0.73	1.44	44.78	40.09	13.74	20.66	1.64	3.00	37.85	54.98

Table 3: Comparison of baseline methods and SCOPE on the S3DIS dataset under the IFS-PCS setting with

K{=}5

and

K{=}1

. We report mIoU, mIoU-B, mIoU-N, HM, mIoU-I, and FPP. All metrics are higher is better except FPP. Best results are highlighted in bold .

Figure 3: Incremental performance on S3DIS and ScanNet for

t{=}0

t{=}3

under

K{=}5

(left) and

K{=}1

(right), respectively. Curves show the evolution of mIoU across incremental stages.

Figure 4: Qualitative comparison with competing methods from $t{=}0$ to $t{=}3$ . The colour palette (right) denotes semantic classes, and dotted separators indicate newly introduced classes at each incremental stage (top to bottom).

Variant	mIoU	mIoU-N	HM	mIoU-I	FPP $\downarrow$
Support Set Only (GW [ 46 ] )	34.27	16.88	23.94	37.67	1.49
+ CPR (mean)	35.68	22.12	28.91	38.02	1.50
+ APE (full)	36.52	23.86	30.38	38.91	1.27

Table 4: Comprehensive ablation study on ScanNet (

K=5

Figure 5: Hyperparameter sensitivity on ScanNet in terms of N-IoU and HM, obtained by varying one parameter at a time while keeping the others fixed (

\tau=0.75

R=20

\lambda=0.5

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Paradigm	Few-shot	Incremental	Prototype-based	Key Idea
FT (Naïve Finetuning)	–	✗	✓	✗	Direct fine-tuning on new tasks without forgetting control
LwF [ 24 ]	CI	✗	✓	✗	Knowledge distillation to preserve previous predictions
EWC [ 21 ]	CI	✗	✓	✗	Fisher-information regularisation to protect important weights
GUA [ 49 ]	CI	✗	✓	✗	Geometry and uncertainty-aware regularisation for PCS
CLIMB-3D [ 39 ]	CI	✗	✓	✗	Handles long-tail distributions in incremental 3D segmentation
AttMPTI [ 53 ]	FS	✓	✗	✓	Multi-prototype inference with label propagation

Appendix

Table A1 : Overview of competitor methods used in our evaluation. Methods span, few-shot learning (FS), generalised few-shot learning (GFS), class-incremental learning (CI), and incremental few-shot learning (IFS) paradigms.

Datasets	Base Classes ( $\bm{C}^{\rm b}$ )	Novel Classes ( $\bm{C}^{\rm n}$ )
S3DIS	Wall, Ceiling, Floor, Clutter, Bookcase, Door, Chair	Beam, Column, Window, Table, Sofa, Board
ScanNet	Refrigerator, Desk, Curtain, Sofa, Bookshelf, Bed, Table, Otherfurniture, Window, Cabinet, Door, Chair, Unannotated, Floor, Wall	Sink, Toilet, Bathtub, Shower Curtain, Picture, Counter

Appendix

Table B2 : Overview of the base and novel class partitions for S3DIS and ScanNet. Following the protocol in the main paper, the majority classes form the base set

\bm{C}^{\rm b}

, while the six least-represented categories constitute the novel set

\bm{C}^{\rm n}

introduced incrementally.

Method	mIoU	mIoU-B	mIoU-N	HM	mIOU-I	FPP
PIFS	25.39	34.81	3.43	6.24	33.11	8.74
+ SC	27.36	34.94	4.93	8.64	34.03	8.61
CAPL	31.73	39.01	14.75	21.36	34.55	-0.65
+ SC	32.97	39.08	18.70	25.25	35.19	-0.72
GW	34.27	41.72	16.88	23.94	37.67	1.49
Ours	36.52	41.94	23.86	30.38	38.91	1.27

Appendix

Table D3 : Effect of adding Scene Contextualisation (SC) to prototype-based IFS-PCS methods on ScanNet (5-shot). SC consistently improves novel-class performance (mIoU-N), incremental-stage accuracy (mIOU-I), and harmonic balance (HM) across all baselines. Applying SC to GW yields our full model ( Ours ), which achieves the best overall performance.

Method	Task	Base Mean	Novel Mean	Novel Classes ( $\bm{C}^{\rm n}$ )
Method	Task	Base Mean	Novel Mean	Sink	Toilet	Bathtub	Shower Curtain	Picture	Counter
GW [ 46 ]	1	40.92	13.45	8.96	17.94	–	–	–	–
	2	41.01	17.16	9.38	21.63	16.64	21.00	–	–
	3	41.72	16.88	9.78	22.54	18.16	22.77	13.78	14.23
SCOPE (Ours)	1	41.02	19.29	12.70	25.88	–	–	–	–
	2	41.15	25.10	12.90	32.33	26.05	29.13	–	–

Appendix

Table E4 : Per-class mIoU across incremental stages on the ScanNet 15B–2I setting. “Mean” denotes the average over base or novel classes. Dashes indicate classes not yet introduced.

Figure E1 : Qualitative comparison of SCOPE with competing baselines from task $t{=}1$ to $t{=}3$ . The colour palette (right) denotes semantic classes, while dotted separators indicate the introduction of new classes at each incremental stage.

Few-Shot Semantic Segmentation

Few-shot semantic segmentation labels pixels in new image classes using only a small set of annotated examples.

See all 32 papers

Learning Hierarchical Orthogonal Prototypes for Generalized Few-Shot 3D Point Cloud Segmentation
Mar 20, 2026Yifei Zhao, Fanyu Zhao, Zhongyuan Zhang +33D Point Cloud Segmentation 3D Semantic Segmentation
College of Computer Science and Artificial Intelligence, Fudan University, Shanghai, China
Few to Big: Prototype Expansion Network via Diffusion Learner for Point Cloud Few-shot Semantic Segmentation
Sep 16, 2025Qianguang Zhao, Dongli Wang, Yan Zhou +23D Point Cloud Segmentation Prototype Learning
School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China · School of Automation and Electronics Information, Xiangtan University, Xiangtan 411105, China · School of Automation, Shanghai Jiao Tong University, Shanghai 200232, China
Uncertainty-aware Prototype Learning with Variational Inference for Few-shot Point Cloud Segmentation
Mar 20, 2026Yifei Zhao, Fanyu Zhao, Yinsheng LiFew-Shot Semantic Segmentation 3D Point Cloud Segmentation
Fudan University, Shanghai, China

Mar 20, 2026J/K move · Enter open · S save

Learning Hierarchical Orthogonal Prototypes for Generalized Few-Shot 3D Point Cloud Segmentation

Yifei Zhao, Fanyu Zhao, Zhongyuan Zhang, Shengtang Wu+2

College of Computer Science and Artificial Intelligence, Fudan University, Shanghai, China

3D Point Cloud Segmentation 3D Semantic Segmentation Few-Shot Semantic Segmentation Prototype Learning

Open paper AskPDF arXiv

Generalized few-shot 3D point cloud segmentation aims to adapt to novel classes from only a few annotations while maintaining strong performance on base classes, but this remains challenging due to the inherent stability-plasticity trade-off: adapting to novel classes can interfere with shared representations and cause base-class forgetting. We present HOP3D, a unified framework that learns hierarchical orthogonal prototypes with an entropy-based few-shot regularizer to enable robust novel-class adaptation without degrading base-class performance. HOP3D introduces hierarchical orthogonalization that decouples base and novel learning at both the gradient and representation levels, effectively mitigating base-novel interference. To further enhance adaptation under sparse supervision, we incorporate an entropy-based regularizer that leverages predictive uncertainty to refine prototype learning and promote balanced predictions. Extensive experiments on ScanNet200 and ScanNet++ demonstrate that HOP3D consistently outperforms state-of-the-art baselines under both 1-shot and 5-shot settings. The code is available at https://fdueblab-hop3d.github.io/.

Abstract

Few-Shot Semantic Segmentation

Explore similar work

Learning Hierarchical Orthogonal Prototypes for Generalized Few-Shot 3D Point Cloud Segmentation

Few to Big: Prototype Expansion Network via Diffusion Learner for Point Cloud Few-shot Semantic Segmentation

Uncertainty-aware Prototype Learning with Variational Inference for Few-shot Point Cloud Segmentation