Incomplete Multi-View Multi-Label Classification via Shared Codebook and Fused-Teacher Self-Distillation

Xu Yan; Jun Yin; Shiliang Sun; Minghua Wan

cs.CVApr 5, 2026

Incomplete Multi-View Multi-Label Classification via Shared Codebook and Fused-Teacher Self-Distillation

Authors: Xu Yan, Jun Yin, Shiliang Sun, Minghua Wan

Organizations: College of Information Engineering, Shanghai Maritime University · School of Automation and Intelligent Sensing, Shanghai Jiao Tong University

Abstract

Although multi-view multi-label learning has been extensively studied, research on the dual-missing scenario, where both views and labels are incomplete, remains largely unexplored. Existing methods mainly rely on contrastive learning or information bottleneck theory to learn consistent representations under missing-view conditions, but loss-based alignment without explicit structural constraints limits the ability to capture stable and discriminative shared semantics. To address this issue, we introduce a more structured mechanism for consistent representation learning: we learn discrete consistent representations through a multi-view shared codebook and cross-view reconstruction, which naturally align different views within the limited shared codebook embeddings and reduce feature redundancy. At the decision level, we design a weight estimation method that evaluates the ability of each view to preserve label correlation structures, assigning weights accordingly to enhance the quality of the fused prediction. In addition, we introduce a fused-teacher self-distillation framework, where the fused prediction guides the training of view-specific classifiers and feeds the global knowledge back into the single-view branches, thereby enhancing the generalization ability of the model under missing-label conditions. The effectiveness of our proposed method is thoroughly demonstrated through extensive comparative experiments with advanced methods on five benchmark datasets. Code is available at https://github.com/xuy11/SCSD.

Figures & tables

Figure 1: The main framework of SCSD. The upper part represents the framework of multi-view consistent discrete representation learning, while the lower part represents the framework of multi-view prediction fusion and self-distillation.

Dataset	$c$	$c/n$	$n$	$m$
Corel5k	260	3.396	4999	6
Pascal07	20	1.465	9963	6
Espgame	268	4.686	20770	6
Iaprtc12	291	5.719	19627	6
Mirflickr	38	4.716	25000	6

Table 1: The summary statistics of different datasets are presented, where

c

denotes the number of classes,

c/n

denotes the average number of positive labels per sample,

n

denotes the number of samples, and

m

denotes the number of views.

Dataset	Metric	iMvWL	NAIM3L	DDINet	DICNet	MTD	SIP	RANK	DRLS	SCSD
	Sources	IJCAI’18	TPAMI’22	TNNLS’23	AAAI’23	NeurIPS’23	ICML’24	TPAMI’25	CVPR’25	—
Corel5k	AP	$0.283_{0.008}$	$0.309_{0.004}$	$0.360_{0.009}$	$0.378_{0.004}$	$0.413_{0.007}$	$0.416_{0.015}$	$0.425_{0.009}$	$\underline{0.433}_{0.008}$	$\pagecolor{gray!25}\boldsymbol{0.447}_{0.010}$
	1-HL	$0.978_{0.000}$	$\underline{0.987}_{0.000}$	$\underline{0.987}_{0.000}$	$\underline{0.987}_{0.000}$	$\boldsymbol{0.988}_{0.000}$	$\boldsymbol{0.988}_{0.000}$	$\boldsymbol{0.988}_{0.000}$	$\boldsymbol{0.988}_{0.000}$	$\pagecolor{gray!25}\boldsymbol{0.988}_{0.000}$
	1-RL	$0.865_{0.005}$	$0.878_{0.002}$	$0.865_{0.005}$	$0.877_{0.004}$	$0.892_{0.004}$	$0.910_{0.003}$	$0.913_{0.003}$	$\underline{0.916}_{0.002}$	$\pagecolor{gray!25}\boldsymbol{0.920}_{0.002}$
	AUC	$0.868_{0.005}$	$0.881_{0.002}$	$0.868_{0.005}$	$0.881_{0.003}$	$0.895_{0.004}$	$0.912_{0.003}$	$0.915_{0.003}$	$\underline{0.918}_{0.002}$	$\pagecolor{gray!25}\boldsymbol{0.923}_{0.003}$
	1-OE	$0.311_{0.015}$	$0.350_{0.009}$	$0.437_{0.012}$	$0.464_{0.012}$	$0.491_{0.010}$	$0.492_{0.018}$	$0.490_{0.014}$	$\underline{0.509}_{0.019}$	$\pagecolor{gray!25}\boldsymbol{0.526}_{0.018}$

Table 2: The results under the setting of 50% missing views, 50% missing labels, and 70% training data are reported. The table lists mean and standard deviation (bottom-right). Ave.R denotes the average rank across metrics. Bold numbers indicate the best results, and underlined numbers indicate the second-best.

Panel of Figure 2 — Figure 2: The radar charts are based on results with complete views, complete labels, and 70% training data, covering nine methods, five datasets, and six metrics. In each chart, the center denotes the worst result and the vertex denotes the best.

Panel of Figure 3 — Figure 3: The parameter sensitivity analysis of the SCSD model is conducted under the setting of 50% missing views, 50% missing labels, and 70% training data.

Method	Corel5k			Pascal07
Method	AP	1-RL	AUC	AP	1-RL	AUC
SCSD w/o $\mathcal{L}_{dis}$	$0.376$	$0.882$	$0.884$	$0.560$	$0.834$	$0.855$
SCSD w/o $\mathcal{L}_{dis\_KL}$	$0.411$	$0.906$	$0.909$	$\underline{0.572}$	$0.843$	$\underline{0.864}$
SCSD w/o $\mathcal{L}_{rec}$	$0.439$	$0.916$	$0.919$	$0.560$	$0.839$	$0.860$
SCSD	$\pagecolor{gray!25}\boldsymbol{0.447}$	$\pagecolor{gray!25}\boldsymbol{0.920}$	$\pagecolor{gray!25}\boldsymbol{0.923}$	$\pagecolor{gray!25}\boldsymbol{0.578}$	$\pagecolor{gray!25}\boldsymbol{0.846}$	$\pagecolor{gray!25}\boldsymbol{0.866}$
SCSD w/o VQ	$0.430$	$0.914$	$0.916$	$0.565$	$0.841$	$0.860$

Table 3: The ablation results on two datasets under the setting of 50% missing views, 50% missing labels, and 70% training data are reported. Here, “w/o” denotes “without”. The bold numbers indicate the best results, while the underlined numbers indicate the second-best results.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 4 — Figure 4: The experimental results of the SCSD model under different view-missing rates, different label-missing rates, and different training set proportions are reported. The figure presents two datasets and three evaluation metrics.

Panel of Figure 5 — Figure 5: An additional parameter sensitivity analysis of the SCSD model is conducted under the setting of 50% missing views, 50% missing labels, and 70% training data.

Figure 6: The codebook utilization of the SCSD method is reported under the training setting of 50% missing views, 50% missing labels, and 70% training data, covering all five datasets.

Multi-Label Classification

Multi-label classification assigns several potentially related tags to each example, from medical findings and topics to objects and user interests.

See all 137 papers

When Semantically Consistent Encoding Meets View-Label Heterogeneity Modeling: A Unified Framework for Incomplete Multi-View Multi-Label Learning
Sep 7, 2026Chengliang Liu, Bo Li, Bob Zhang +3Multi-View Consistency Multi-Label Classification
PAMI Research Group, Department of Artificial Intelligence, University of Macau, Macau · PAMI Research Group, Department of Artificial Intelligence, and Centre for Artificial General Intelligence, Institute of Artificial Intelligence and Brain Sciences, University of Macau, Macau · School of Computer Science and Technology, Beijing Institute of Technology, Beijing 100811, China +2
No Data Wasted: A Semi-supervised Generative Model for Incomplete Multi-view Data Integration with Missing Labels
Aug 15, 2025Yiyang Shen, Weiran WangLearning with Missing Data Multi-View Learning
Department of Computer Science, University of Iowa, Iowa City, Iowa, USA
SPORT: Structure-Aware Prototype Disentanglement for Incomplete Multi-View Clustering
Jul 11, 2026Yaoyuan Guo, Zhibin Gu, Songhe Feng +2Multi-View Consistency Disentangled Representation Learning
College of Computer and Cyberspace Security, Hebei Normal University, Shijiazhuang 050024, China · Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, China · Key Laboratory of Tibetan Information Processing, Ministry of Education, Qinghai Normal University, Xining 810008, China +2

Sep 7, 2026J/K move · Enter open · S save

When Semantically Consistent Encoding Meets View-Label Heterogeneity Modeling: A Unified Framework for Incomplete Multi-View Multi-Label Learning

Chengliang Liu, Bo Li, Bob Zhang, Yanghao Zhou+2

PAMI Research Group, Department of Artificial Intelligence, University of Macau, Macau · PAMI Research Group, Department of Artificial Intelligence, and Centre for Artificial General Intelligence, Institute of Artificial Intelligence and Brain Sciences, University of Macau, Macau · School of Computer Science and Technology, Beijing Institute of Technology, Beijing 100811, China+2

Multi-View Consistency Multi-Label Classification Multi-View Learning

Open paper AskPDF arXiv

Incomplete multi-view multi-label learning requires not only robust semantic aggregation from partially observed views, but also label-aware exploitation of view-specific evidence. Existing approaches usually emphasize either shared representation learning or decision-level fusion. The former improves robustness against missing views, yet tends to compress label-discriminative view-specific cues into a single latent representation. The latter preserves individual view predictions, but often relies on fixed or globally learned fusion weights, ignoring that different labels of different instances may require different views. To address these limitations, this paper presents V2L, a unified representation-decision framework for incomplete multi-view multi-label classification. On the representation side, V2L constructs semantically consistent variational posteriors from incomplete views through a perturbation-aware encoding mechanism, which provides a stable shared semantic basis. On the decision side, V2L introduces an active view-label relevance modeling strategy that estimates instance-wise and label-wise view contributions, allowing each label prediction to adaptively select useful view-specific evidence. From the perspective of model architecture, these two important strategies are integrated into a unified framework through a hybrid fusion architecture, simultaneously meeting the requirements of cross-view semantic consistency and representational complementarity. Extensive experiments under both incomplete and complete settings show that V2L achieves leading performance on five benchmarks. Code is available at: https://github.com/justsmart/V2L.

Abstract

Multi-Label Classification

Explore similar work

When Semantically Consistent Encoding Meets View-Label Heterogeneity Modeling: A Unified Framework for Incomplete Multi-View Multi-Label Learning

No Data Wasted: A Semi-supervised Generative Model for Incomplete Multi-view Data Integration with Missing Labels

SPORT: Structure-Aware Prototype Disentanglement for Incomplete Multi-View Clustering