Prompt-based learning has shown promise in visual-language tracking (VLT), yet existing methods often rely on either explicit or implicit prompting alone, limiting fine-grained cross-modal alignment. Moreover, Low-Rank Adaptation (LoRA) -based fine-tuning in prior work typically focuses on visual-only adaptation, overlooking language semantics. To address these issues, we propose a unified VLT framework that integrates Explicit-Implicit Prompt Injection (EIPI) and Semantic-Guided Latent LoRA (SGLL). EIPI introduces semantic prompts to facilitate robust and context-sensitive target modeling through two pathways. The explicit prompts are constructed by interact between multi-modal target representations with the search region, while implicit prompts are learned from linguistic features via a lightweight bottleneck network. Then, SGLL extends standard LoRA by introducing learnable queries in the latent space, allowing residual modulation based on language-visual semantics without retraining the full model. This dual design yields a parameter-efficient tracker with strong cross-modal adaptability. Extensive experiments show our method outperforms prior prompt-based approaches while maintaining high efficiency.

Existing RGB-T salient object detection methods primarily rely on the learning mechanism of neural networks to perform implicit cross-modal feature alignment, aiming to achieve complementary fusion of modal features. However, this implicit feature alignment method has two main limitations: first, it is prone to causing loss of the salient object’s structural information; second, it may lead to abnormal activation responses that are not related to the object. To address the above issues, we propose the innovative Explicit Semantic Alignment (ESA) framework and design the Explicit Semantic Alignment Network for RGB-T Salient Object Detection with Hierarchical Cross-Modal Fusion (ESANet). Specifically, we design a Saliency-Aware Refinement Module (SARM), which fuses high-level semantic features with mid-level spatial details through cross-aggregation and the dynamic integration module to achieve bidirectional interaction and adaptive fusion of cross-modal features. It also utilizes a cross-modal multi-head attention mechanism to generate fine-grained shared semantic information. Subsequently, the Cross-Modal Feature Alignment Module (CFAM) introduces a window-based attention propagation mechanism, which enforces consistency in scene understanding between RGB and thermal modalities by using shared semantics as an alignment constraint. Finally, the Semantic-Guided Edge Sharpening Module (SESM) combines shared semantics with a weight enhancement strategy to optimize the consistency of shallow cross-modal feature distributions. Experimental results demonstrate that ESANet significantly outperforms existing state-of-the-art RGB-T salient object detection methods on three public datasets, validating its excellent performance in salient object detection tasks. Our code will be released at https://github.com/whklearn/ESANet.git .

Abstract Existing RGB‐D salient object detection methods mainly rely on a symmetric two‐stream Convolutional Neural Network (CNN)‐based network to extract RGB and depth channel features separately. However, there are two problems with the symmetric conventional network structure: first, the ability of CNN in learning global contexts is limited; second, the symmetric two‐stream structure ignores the inherent differences between modalities. In this study, a Transformer‐based asymmetric network is proposed to tackle the issues mentioned above. The authors employ the powerful feature extraction capability of Transformer to extract global semantic information from RGB data and design a lightweight CNN backbone to extract spatial structure information from depth data without pre‐training. The asymmetric hybrid encoder effectively reduces the number of parameters in the model while increasing speed without sacrificing performance. Then, a cross‐modal feature fusion module which enhances and fuses RGB and depth features with each other is designed. Finally, the authors add edge prediction as an auxiliary task and propose an edge enhancement module to generate sharper contours. Extensive experiments demonstrate that our method achieves superior performance over 14 state‐of‐the‐art RGB‐D methods on six public datasets. The code of the authors will be released at https://github.com/lc012463/TANet .

In this work, we develop a real-time tracking algorithm with a lightweight deep neural network. The contributions of this work mainly include two aspects. First, we reformulate the discriminative correlation filter (DCF) based tracker as a fully convolutional neural network and design an effective end-to-end tracking framework. Second, we build our tracker with a pruned convolutional neural network, which is trained by a mutual learning approach to further improve the location accuracy. The proposed tracking algorithm can track objects at 60 FPS. Extensive experiments on OTB2013, OTB2015 and VOT2017 Real-time demonstrate that the proposed tracker performs favorably against state-of-the-art methods.

Effective integration of contextual information is crucial for salient object detection. To achieve this, most existing methods based on 'skip' architecture mainly focus on how to integrate hierarchical features of Convolutional Neural Networks (CNNs). They simply apply concatenation or element-wise operation to incorporate high-level semantic cues and low-level detailed information. However, this can degrade the quality of predictions because cluttered and noisy information can also be passed through. To address this problem, we proposes a global Recurrent Localization Network (RLN) which exploits contextual information by the weighted response map in order to localize salient objects more accurately. Particularly, a recurrent module is employed to progressively refine the inner structure of the CNN over multiple time steps. Moreover, to effectively recover object boundaries, we propose a local Boundary Refinement Network (BRN) to adaptively learn the local contextual information for each spatial position. The learned propagation coefficients can be used to optimally capture relations between each pixel and its neighbors. Experiments on five challenging datasets show that our approach performs favorably against all existing methods in terms of the popular evaluation metrics.

Fully convolutional networks (FCN) has significantly improved the performance of many pixel-labeling tasks, such as semantic segmentation and depth estimation. However, it still remains nontrivial to thoroughly utilize the multilevel convolutional feature maps and boundary information for salient object detection. In this letter, we propose a novel FCN framework to integrate multilevel convolutional features recurrently with the guidance of object boundary information. First, a deep convolutional network is used to extract multilevel feature maps and separately aggregate them into multiple resolutions, which can be used to generate coarse saliency maps. Meanwhile, another boundary information extraction branch is proposed to generate boundary features. Finally, an attention-based feature fusion module is designed to fuse boundary information into salient regions to achieve accurate boundary inference and semantic enhancement. The final saliency maps are the combination of the predicted boundary maps and integrated saliency maps, which are more closer to the ground truths. Experiments and analysis on four large-scale benchmarks verify that our framework achieves new state-of-the-art results.

We propose a novel unsupervised game-theoretic salient object detection algorithm that does not require labeled training data. First, saliency detection problem is formulated as a non-cooperative game, hereinafter referred to as Saliency Game, in which image regions are players who choose to be "background" or "foreground" as their pure strategies. A payoff function is constructed by exploiting multiple cues and combining complementary features. Saliency maps are generated according to each region's strategy in the Nash equilibrium of the proposed Saliency Game. Second, we explore the complementary relationship between color and deep features and propose an Iterative Random Walk algorithm to combine saliency maps produced by the Saliency Game using different features. Iterative random walk allows sharing information across feature spaces, and detecting objects that are otherwise very hard to detect. Extensive experiments over 6 challenging datasets demonstrate the superiority of our proposed unsupervised algorithm compared to several state of the art supervised algorithms.