Zero-shot skeleton action recognition endeavors to classify novel action categories by transferring previously learned seen skeleton-semantic priors to unseen categories. However, current methods struggle to distinguish highly similar action categories, primarily due to the coarse-grained cross-modal alignment and non-discriminative representation space. To address these issues, we proposeSTAR++, a novel framework that aligns skeleton and semantics in a fine-grained and conditional manner. The key idea is to first establish region-level correspondences between body parts and semantic cues, and then utilize these local alignments to inform a global alignment process. This design is inspired by human visual cognition, which first attends to crucial local details before perceiving the broader scene. Concretely, we refine both skeleton and semantic representations with a dual-prompt attention mechanism driven by the structural decomposition of the human body and side information generated by a large language model (LLM). This encourages skeleton representations to be more compact within each class and semantic embeddings to be more separable across classes, which helps resolve ambiguity between highly similar actions and provides better interpretability of how unseen actions are perceived. Furthermore, we construct a region-aware holistic fusion module that aggregates these fine-grained features into a unified representation, yielding more discriminative holistic representations. Finally, the global alignment is conditioned on region-aware semantics feedback derived from fine-grained alignment, forming a conditional process that achieves more effective cross-modal alignment. Extensive experiments on four mainstream benchmarks demonstrate that our method achieves state-of-the-art performance in the zero-shot learning (ZSL) and generalized zero-shot learning (GZSL) settings.

With the increasing attention to pre-trained vision-language models (VLMs), \eg, CLIP, substantial efforts have been devoted to many downstream tasks, especially in test-time adaptation (TTA). However, previous works focus on learning prototypes only in the textual modality while overlooking the ambiguous semantics in class names. These ambiguities lead to textual prototypes that are insufficient to capture visual concepts, resulting in limited performance. To address this issue, we introduce \textbf{ProtoMM}, a training-free framework that constructs multimodal prototypes to adapt VLMs during the test time. By viewing the prototype as a discrete distribution over the textual descriptions and visual particles, ProtoMM has the ability to combine the multimodal features for comprehensive prototype learning. More importantly, the visual particles are dynamically updated as the testing stream flows. This allows our multimodal prototypes to continually learn from the data, enhancing their generalizability in unseen scenarios. In addition, we quantify the importance of the prototypes and test images by formulating their semantic distance as an optimal transport problem. Extensive experiments on 15 zero-shot benchmarks demonstrate the effectiveness of our method, achieving a 1.03\% average accuracy improvement over state-of-the-art methods on ImageNet and its variant datasets.

Compositional Zero-Shot Learning (CZSL) aims to recognize novel state-object compositions by leveraging the shared knowledge of their primitive components. Despite considerable progress, effectively calibrating the bias between semantically similar multimodal representations, as well as generalizing pre-trained knowledge to novel compositional contexts, remains an enduring challenge. In this paper, our interest is to revisit the conditional transport (CT) theory and its homology to the visual-semantics interaction in CZSL and further, propose a novel Trisets Consistency Alignment framework (dubbed TsCA) that well-addresses these issues. Concretely, we utilize three distinct yet semantically homologous sets, i.e., patches, primitives, and compositions, to construct pairwise CT costs to minimize their semantic discrepancies. To further ensure the consistency transfer within these sets, we implement a cycle-consistency constraint that refines the learning by guaranteeing the feature consistency of the self-mapping during transport flow, regardless of modality. Moreover, we extend the CT plans to an open-world setting, which enables the model to effectively filter out unfeasible pairs, thereby speeding up the inference as well as increasing the accuracy. Extensive experiments are conducted to verify the effectiveness of the proposed method. The code is available at https://github.com/keepgoingjkg/TsCA.

Conditional dependency present one of the trickiest problems in Compositional Zero-Shot Learning, leading to significant property variations of the same state (object) across different objects (states). To address this problem, existing approaches often adopt either all-to-one or one-to-one representation paradigms. However, these extremes create an imbalance in the seesaw between transferability and discriminability, favoring one at the expense of the other. Comparatively, humans are adept at analogizing and reasoning in a hierarchical clustering manner, intuitively grouping categories with similar properties to form cohesive concepts. Motivated by this, we propose Homogeneous Group Representation Learning (HGRL), a new perspective formulates state (object) representation learning as multiple homogeneous sub-group representation learning. HGRL seeks to achieve a balance between semantic transferability and discriminability by adaptively discovering and aggregating categories with shared properties, learning distributed group centers that retain group-specific discriminative features. Our method integrates three core components designed to simultaneously enhance both the visual and prompt representation capabilities of the model. Extensive experiments on three benchmark datasets validate the effectiveness of our method. Code is available at https://github.com/zjrao/HGRL.

For downstream applications of vision-language pre-trained models, there has been significant interest in constructing effective prompts. Existing works on prompt engineering, which either require laborious manual designs or optimize the prompt tuning as a point estimation problem, may fail to describe diverse characteristics of categories and limit their applications. We introduce a Bayesian probabilistic resolution to prompt tuning, where the label-specific stochastic prompts are generated hierarchically by first sampling a latent vector from an underlying distribution and then employing a lightweight generative model. Importantly, we semantically regularize the tuning process by minimizing the statistic distance between the visual patches and linguistic prompts, which pushes the stochastic label representations to faithfully capture diverse visual concepts, instead of overfitting the training categories. We evaluate the effectiveness of our approach on four tasks: few-shot image recognition, base-to-new generalization, dataset transfer learning, and domain shifts. Extensive results on over 15 datasets show promising transferability and generalization performance of our proposed model, both quantitatively and qualitatively.

Multi-label image classification is a prediction task that aims to identify more than one label from a given image. This paper considers the semantic consistency of the latent space between the visual patch and linguistic label domains and introduces the conditional transport (CT) theory to bridge the acknowledged gap. While recent cross-modal attention-based studies have attempted to align such two representations and achieved impressive performance, they required carefully-designed alignment modules and extra complex operations in the attention computation. We find that by formulating the multi-label classification as a CT problem, we can exploit the interactions between the image and label efficiently by minimizing the bidirectional CT cost. Specifically, after feeding the images and textual labels into the modality-specific encoders, we view each image as a mixture of patch embeddings and a mixture of label embeddings, which capture the local region features and the class prototypes, respectively. CT is then employed to learn and align those two semantic sets by defining the forward and backward navigators. Importantly, the defined navigators in CT distance model the similarities between patches and labels, which provides an interpretable tool to visualize the learned prototypes. Extensive experiments on three public image benchmarks show that the proposed model consistently outperforms the previous methods.

Advancements in prompt tuning of vision-language models have underscored their potential in enhancing open-world visual concept comprehension. However, prior works only primarily focus on single-mode (only one prompt for each modality) and holistic level (image or sentence) semantic alignment, which fails to capture the sample diversity, leading to sub-optimal prompt discovery. To address the limitation, we propose a multi-mode token-level tuning framework that leverages the optimal transportation to learn and align a set of prompt tokens across modalities. Specifically, we rely on two essential factors: 1) multi-mode prompts discovery, which guarantees diverse semantic representations, and 2) token-level alignment, which helps explore fine-grained similarity. Consequently, the similarity can be calculated as a hierarchical transportation problem between the modality-specific sets. Extensive experiments on popular image recognition benchmarks show the superior generalization and few-shot abilities of our approach. The qualitative analysis demonstrates that the learned prompt tokens have the ability to capture diverse visual concepts.

We propose a Bayesian generative model for incorporating prior domain knowledge into hierarchical topic modeling. Although embedded topic models (ETMs) and its variants have gained promising performance in text analysis, they mainly focus on mining word co-occurrence patterns, ignoring potentially easy-to-obtain prior topic hierarchies that could help enhance topic coherence. While several knowledge-based topic models have recently been proposed, they are either only applicable to shallow hierarchies or sensitive to the quality of the provided prior knowledge. To this end, we develop a novel deep ETM that jointly models the documents and the given prior knowledge by embedding the words and topics into the same space. Guided by the provided domain knowledge, the proposed model tends to discover topic hierarchies that are organized into interpretable taxonomies. Moreover, with a technique for adapting a given graph, our extended version allows the structure of the prior knowledge to be fine-tuned to match the target corpus. Extensive experiments show that our proposed model efficiently integrates the prior knowledge and improves both hierarchical topic discovery and document representation.