arXiv:2511.07780v1 Announce Type: new 
Abstract: Cross-modal hashing (CMH) facilitates efficient retrieval across different modalities (e.g., image and text) by encoding data into compact binary representations. While recent methods have achieved remarkable performance, they often rely heavily on fully annotated datasets, which are costly and labor-intensive to obtain. In real-world scenarios, particularly in multi-label datasets, label noise is prevalent and severely degrades retrieval performance. Moreover, existing CMH approaches typically overlook the partial semantic overlaps inherent in multi-label data, limiting their robustness and generalization. To tackle these challenges, we propose a novel framework named Semantic-Consistent Bidirectional Contrastive Hashing (SCBCH). The framework comprises two complementary modules: (1) Cross-modal Semantic-Consistent Classification (CSCC), which leverages cross-modal semantic consistency to estimate sample reliability and reduce the impact of noisy labels; (2) Bidirectional Soft Contrastive Hashing (BSCH), which dynamically generates soft contrastive sample pairs based on multi-label semantic overlap, enabling adaptive contrastive learning between semantically similar and dissimilar samples across modalities. Extensive experiments on four widely-used cross-modal retrieval benchmarks validate the effectiveness and robustness of our method, consistently outperforming state-of-the-art approaches under noisy multi-label conditions.

تم اقتراح إطار جديد يسمى 'التشفير الثنائي المتباين المتسق دلاليًا' (SCBCH) لتحسين أداء الاسترجاع عبر الأنماط في مجموعات البيانات متعددة التسميات المليئة بالضوضاء. يتناول هذا الإطار التحديات المتعلقة بضوضاء التسميات والتداخلات الدلالية الجزئية، التي غالبًا ما تؤدي إلى تدهور دقة الاسترجاع. يتضمن SCBCH وحدتين: التصنيف الدلالي المتسق عبر الأنماط والتشفير الثنائي المتباين الناعم، وقد أظهر أداءً متفوقًا في التجارب على أربعة معايير مستخدمة على نطاق واسع.

Se ha propuesto un nuevo marco llamado Semantic-Consistent Bidirectional Contrastive Hashing (SCBCH) para mejorar el rendimiento de la recuperación cross-modal en conjuntos de datos multi-etiqueta ruidosos. Este marco aborda los desafíos del ruido de etiquetas y las superposiciones semánticas parciales, que a menudo degradan la precisión de la recuperación. SCBCH incluye dos módulos: Clasificación Semántica Consistente Cross-Modal y Hachado Contrastivo Bidireccional Suave, y ha mostrado un rendimiento superior en experimentos en cuatro benchmarks ampliamente utilizados.

Un nouveau cadre appelé Semantic-Consistent Bidirectional Contrastive Hashing (SCBCH) a été proposé pour améliorer la performance de récupération cross-modal dans des ensembles de données multi-label bruyants. Ce cadre aborde les défis du bruit d'étiquetage et des chevauchements sémantiques partiels, qui dégradent souvent la précision de récupération. SCBCH comprend deux modules : la classification sémantique cross-modal cohérente et le hachage contrastif bidirectionnel souple, et a montré des performances supérieures lors d'expériences sur quatre benchmarks largement utilisés.

A new framework called Semantic-Consistent Bidirectional Contrastive Hashing (SCBCH) has been proposed to enhance cross-modal retrieval performance in noisy multi-label datasets. This framework addresses the challenges of label noise and partial semantic overlaps, which often degrade retrieval accuracy. SCBCH includes two modules: Cross-modal Semantic-Consistent Classification and Bidirectional Soft Contrastive Hashing, and has shown superior performance in experiments on four widely-used benchmarks.

Semantic-Consistent Bidirectional Contrastive Hashing for Noisy Multi-Label Cross-Modal Retrieval

arXiv:2601.08257v1 Announce Type: new 
Abstract: Unsupervised feature selection aims to identify a compact subset of features that captures the intrinsic structure of data without supervised label. Most existing studies evaluate the performance of methods using the single-label dataset that can be instantiated by selecting a label from multi-label data while maintaining the original features. Because the chosen label can vary arbitrarily depending on the experimental setting, the superiority among compared methods can be changed with regard to which label happens to be selected. Thus, evaluating unsupervised feature selection methods based solely on single-label accuracy is unreasonable for assessing their true discriminative ability. This study revisits this evaluation paradigm by adopting a multi-label classification framework. Experiments on 21 multi-label datasets using several representative methods demonstrate that performance rankings differ markedly from those reported under single-label settings, suggesting the possibility of multi-label evaluation settings for fair and reliable comparison of unsupervised feature selection methods.

دراسة حديثة نُشرت على arXiv تقيم طرق اختيار الميزات غير الخاضعة للإشراف ضمن إطار تصنيف متعدد التسميات، مشددة على أن التقييمات التقليدية التي تعتمد على دقة تسمية واحدة قد تمثل بشكل خاطئ القدرات التمييزية الحقيقية لهذه الطرق. شملت الأبحاث تجارب على 21 مجموعة بيانات متعددة التسميات، كاشفة عن اختلافات كبيرة في تصنيفات الأداء مقارنة بالتقييمات السابقة.

Un estudio reciente publicado en arXiv evalúa los métodos de selección de características no supervisadas dentro de un marco de clasificación multi-etiqueta, destacando que las evaluaciones tradicionales basadas en la precisión de una sola etiqueta pueden representar incorrectamente las verdaderas capacidades discriminativas de estos métodos. La investigación incluyó experimentos en 21 conjuntos de datos multi-etiqueta, revelando diferencias significativas en los rankings de rendimiento en comparación con evaluaciones anteriores.

Une étude récente publiée sur arXiv évalue les méthodes de sélection de caractéristiques non supervisées dans un cadre de classification multi-label, soulignant que les évaluations traditionnelles basées sur la précision à étiquette unique peuvent mal représenter les véritables capacités discriminatives de ces méthodes. La recherche a impliqué des expériences sur 21 ensembles de données multi-label, révélant des différences significatives dans les classements de performance par rapport aux évaluations précédentes.

A recent study published on arXiv evaluates unsupervised feature selection methods within a multi-label classification framework, highlighting that traditional single-label accuracy assessments may misrepresent the true discriminative capabilities of these methods. The research involved experiments on 21 multi-label datasets, revealing significant differences in performance rankings compared to previous evaluations.

Semantic-Consistent Bidirectional Contrastive Hashing for Noisy Multi-Label Cross-Modal Retrieval

Was this article worth reading? Share it

Ready to build your own newsroom?