Fine-grained image-text alignment is a pivotal challenge in multimodal learning, underpinning key applications such as visual question answering, image captioning, and vision-language navigation. Unlike global alignment, fine-grained alignment requires precise correspondence between localized visual regions and textual tokens, often hindered by noisy attention mechanisms and oversimplified modeling of cross-modal relationships. In this work, we identify two fundamental limitations of existing approaches: the lack of robust intra-modal mechanisms to assess the significance of visual and textual tokens, leading to poor generalization in complex scenes; and the absence of fine-grained uncertainty modeling, which fails to capture the one-to-many and many-to-one nature of region-word correspondences. To address these issues, we propose a unified approach that incorporates significance-aware and granularity-aware modeling and region-level uncertainty modeling. Our method leverages modality-specific biases to identify salient features without relying on brittle cross-modal attention, and represents region features as a mixture of Gaussian distributions to capture fine-grained uncertainty. Extensive experiments on Flickr30K and MS-COCO demonstrate that our approach achieves state-of-the-art performance across various backbone architectures, significantly enhancing the robustness and interpretability of fine-grained image-text alignment.

The feature dispersion, arising from the inherent constraints of data scarcity, has emerged as a prominent challenge in the domain of few-shot learning. In this paper, we propose a novel Self-adaptive Distribution Regularization (SADR) approach, which can adaptively bridge the semantic gaps across distribution patterns and boundaries for learning from limited labeled data. Specifically, the technical core of our SADR approach is to decouple the feature embeddings into two discrete spaces: the intra-class and inter-class distributions, leading to robust and discriminative feature representations in a self-adaptive manner. To achieve meticulous similarity measurements while mitigating redundant feature information, an innovative regularized Brownian Distance Covariance (R-BDC) metric is strategically designed to simultaneously explore both the joint and marginal distributions present among diverse input samples. Extensive experiments on multiple benchmark datasets consistently demonstrate the superiority of our SADR approach over state-of-the-art baselines.

Most recent few-shot learning approaches are based on meta-learning with episodic training. However, prior studies encounter two crucial problems: (1) the presence of inductive bias, and (2) the occurrence of catastrophic forgetting. In this paper, we propose a novel Multi-Level Contrastive Constraints (MLCC) framework, that jointly integrates within-episode learning and across-episode learning into a unified interactive learning paradigm to solve these issues. Specifically, we employ a space-aware interaction modeling scheme to explore the correct inductive paradigms for each class between within-episode similarity/dis-similarity distributions. Additionally, with the aim of better utilizing former prior knowledge, a cross-stage distribution adaption strategy is designed to align the across-episode distributions from different time stages, thus reducing the semantic gap between existing and past prediction distribution. Extensive experiments on multiple few-shot datasets demonstrate the consistent superiority of MLCC approach over the existing state-of-the-art baselines.

We present a new vision-language (VL) pre-training model dubbed Kaleido-BERT , which introduces a novel kaleido strategy for fashion cross-modality representations from transformers. In contrast to random masking strategy of recent VL models, we design alignment guided masking to jointly focus more on image-text semantic relations. To this end, we carry out five novel tasks, i.e., rotation, jigsaw, camouflage, grey-to-color, and blank-to-color for self-supervised VL pre-training at patches of different scale. Kaleido-BERT is conceptually simple and easy to extend to the existing BERT framework, it attains state-of-the-art results by large margins on four downstream tasks, including text retrieval (R@1: 4.03% absolute improvement), image retrieval (R@1: 7.13% abs imv.), category recognition (ACC: 3.28% abs imv.), and fashion captioning (Bleu4: 1.2 abs imv.). We validate the efficiency of Kaleido-BERT on a wide range of e-commerical websites, demonstrating its broader potential in real-world applications.