Abstract:Fine-grained representation is fundamental to species classification based on deep learning, and in this context, cross-modal contrastive learning is an effective method. The diversity of species coupled with the inherent contextual ambiguity of natural language poses a primary challenge in the cross-modal representation alignment of conservation area image data. Integrating cross-modal retrieval tasks with generation tasks contributes to cross-modal representation alignment based on contextual understanding. However, during the contrastive learning process, apart from learning the differences in the data itself, a pair of encoders inevitably learns the differences caused by encoder fluctuations. The latter leads to convergence shortcuts, resulting in poor representation quality and an inaccurate reflection of the similarity relationships between samples in the original dataset within the shared space of features. To achieve fine-grained cross-modal representation alignment, we first propose a residual attention network to enhance consistency during momentum updates in cross-modal encoders. Building upon this, we propose momentum encoding from a multi-task perspective as a bridge for cross-modal information, effectively improving cross-modal mutual information, representation quality, and optimizing the distribution of feature points within the cross-modal shared semantic space. By acquiring momentum encoding queues for cross-modal semantic understanding through multi-tasking, we align ambiguous natural language representations around the invariant image features of factual information, alleviating contextual ambiguity and enhancing model robustness. Experimental validation shows that our proposed multi-task perspective of cross-modal momentum encoders outperforms similar models on standardized image classification tasks and image–text cross-modal retrieval tasks on public datasets by up to 8% on the leaderboard, demonstrating the effectiveness of the proposed method. Qualitative experiments on our self-built conservation area image–text paired dataset show that our proposed method accurately performs cross-modal retrieval and generation tasks among 8142 species, proving its effectiveness on fine-grained cross-modal image–text conservation area image datasets.

Select & Re-Rank: Effectively and Efficiently Matching Multimodal Data with Dynamically Evolving Attention

Embrace Smaller Attention: Efficient Cross-Modal Matching with Dual Gated Attention Fusion

Dynamic Visual Semantic Sub-Embeddings and Fast Re-Ranking

Visual-Semantic Matching by Exploring High-Order Attention and Distraction

Interactive Attention for Semantic Text Matching

Fine-Grained Cross-Modal Semantic Consistency in Natural Conservation Image Data from a Multi-Task Perspective

Multi-level network based on transformer encoder for fine-grained image–text matching

Multi-Modal Memory Enhancement Attention Network for Image-Text Matching

Dual Semantic Relationship Attention Network for Image-Text Matching

Semantic Matching Using an Efficient Multi-task Method

Comateformer: Combined Attention Transformer for Semantic Sentence Matching

Attention-based multimodal image matching

Learning Dual Semantic Relations with Graph Attention for Image-Text Matching

Semantic enhancement and multi-level alignment network for cross-modal retrieval

Matching Images and Text with Multi-modal Tensor Fusion and Re-ranking

Image-Text Matching with Multi-View Attention

Rethinking the Multimodal Correlation of Multimodal Sequential Learning via Generalizable Attentional Results Alignment

Multimodal Remote Sensing Image Matching via Learning Features and Attention Mechanism

Interactive Attention Networks for Semantic Text Matching

Enhanced Distance-Aware Self-Attention and Multi-Level Match for Sentence Semantic Matching