Zhijian Wu

Guangtao Wang, Jun Li, Zhijian Wu et al.

In recent years, deep convolutional neural networks (CNN) have significantly advanced face detection. In particular, lightweight CNNbased architectures have achieved great success due to their lowcomplexity structure facilitating real-time detection tasks. However, current lightweight CNN-based face detectors trading accuracy for efficiency have inadequate capability in handling insufficient feature representation, faces with unbalanced aspect ratios and occlusion. Consequently, they exhibit deteriorated performance far lagging behind the deep heavy detectors. To achieve efficient face detection without sacrificing accuracy, we design an efficient deep face detector termed EfficientFace in this study, which contains three modules for feature enhancement. To begin with, we design a novel cross-scale feature fusion strategy to facilitate bottom-up information propagation, such that fusing low-level and highlevel features is further strengthened. Besides, this is conducive to estimating the locations of faces and enhancing the descriptive power of face features. Secondly, we introduce a Receptive Field Enhancement module to consider faces with various aspect ratios. Thirdly, we add an Attention Mechanism module for improving the representational capability of occluded faces. We have evaluated EfficientFace on four public benchmarks and experimental results demonstrate the appealing performance of our method. In particular, our model respectively achieves 95.1% (Easy), 94.0% (Medium) and 90.1% (Hard) on validation set of WIDER Face dataset, which is competitive with heavyweight models with only 1/15 computational costs of the state-of-the-art MogFace detector.

Guang Yang, Yin Tang, Zhijian Wu et al.

Recent mainstream masked distillation methods function by reconstructing selectively masked areas of a student network from the feature map of its teacher counterpart. In these methods, the masked regions need to be properly selected, such that reconstructed features encode sufficient discrimination and representation capability like the teacher feature. However, previous masked distillation methods only focus on spatial masking, making the resulting masked areas biased towards spatial importance without encoding informative channel clues. In this study, we devise a Dual Masked Knowledge Distillation (DMKD) framework which can capture both spatially important and channel-wise informative clues for comprehensive masked feature reconstruction. More specifically, we employ dual attention mechanism for guiding the respective masking branches, leading to reconstructed feature encoding dual significance. Furthermore, fusing the reconstructed features is achieved by self-adjustable weighting strategy for effective feature distillation. Our experiments on object detection task demonstrate that the student networks achieve performance gains of 4.1% and 4.3% with the help of our method when RetinaNet and Cascade Mask R-CNN are respectively used as the teacher networks, while outperforming the other state-of-the-art distillation methods.

Jingchao Wang, Kaiwen Zhou, Zhijian Wu et al.

Vision-Language Tracking aims to continuously localize objects described by a visual template and a language description. Existing methods, however, are typically limited to local search, making them prone to failures under viewpoint changes, occlusions, and rapid target movements. In this work, we introduce the first global tracking framework based on Multimodal Large Language Models (VPTracker), exploiting their powerful semantic reasoning to locate targets across the entire image space. While global search improves robustness and reduces drift, it also introduces distractions from visually or semantically similar objects. To address this, we propose a location-aware visual prompting mechanism that incorporates spatial priors into the MLLM. Specifically, we construct a region-level prompt based on the target's previous location, enabling the model to prioritize region-level recognition and resort to global inference only when necessary. This design retains the advantages of global tracking while effectively suppressing interference from distracting visual content. Extensive experiments show that our approach significantly enhances tracking stability and target disambiguation under challenging scenarios, opening a new avenue for integrating MLLMs into visual tracking. Code is available at https://github.com/jcwang0602/VPTracker.

Zhourui Zhang, Jun Li, Zhijian Wu et al.

In recent years, current mainstream feature masking distillation methods mainly function by reconstructing selectively masked regions of a student network from the feature maps of a teacher network. In these methods, attention mechanisms can help to identify spatially important regions and crucial object-aware channel clues, such that the reconstructed features are encoded with sufficient discriminative and representational power similar to teacher features. However, previous feature-masking distillation methods mainly address homogeneous knowledge distillation without fully taking into account the heterogeneous knowledge distillation scenario. In particular, the huge discrepancy between the teacher and the student frameworks within the heterogeneous distillation paradigm is detrimental to feature masking, leading to deteriorating reconstructed student features. In this study, a novel dual feature-masking heterogeneous distillation framework termed DFMSD is proposed for object detection. More specifically, a stage-wise adaptation learning module is incorporated into the dual feature-masking framework, and thus the student model can be progressively adapted to the teacher models for bridging the gap between heterogeneous networks. Furthermore, a masking enhancement strategy is combined with stage-wise learning such that object-aware masking regions are adaptively strengthened to improve feature-masking reconstruction. In addition, semantic alignment is performed at each Feature Pyramid Network (FPN) layer between the teacher and the student networks for generating consistent feature distributions. Our experiments for the object detection task demonstrate the promise of our approach, suggesting that DFMSD outperforms both the state-of-the-art heterogeneous and homogeneous distillation methods.

Hongze Zhu, Chen Hu, Jiaxuan Jiang et al.

Vision-Language Models (VLMs) have significantly advanced automated Radiology Report Generation (RRG). However, existing methods implicitly assume high-quality inputs, overlooking the noise and artifacts prevalent in real-world clinical environments. Consequently, current models exhibit severe performance degradation when processing suboptimal images. To bridge this gap, we propose a robust report generation framework explicitly designed for image quality variations. We first introduce an Automated Quality Assessment Agent (AQAA) to identify low-quality samples within the MIMIC-CXR dataset and establish the Low-quality Radiology Report Generation (LRRG) benchmark. To tackle degradation-induced shifts, we propose a novel Dual-loop Training Strategy leveraging bi-level optimization and gradient consistency. This approach ensures the model learns quality-agnostic diagnostic features by aligning gradient directions across varying quality regimes. Extensive experiments demonstrate that our approach effectively mitigates model performance degradation caused by image quality deterioration. The code and data will be released upon acceptance.

Jingchao Wang, Zhijian Wu, Dingjiang Huang et al.

Reference Expression Segmentation (RES) aims to segment image regions specified by referring expressions and has become popular with the rise of multimodal large models (MLLMs). While MLLMs excel in semantic understanding, their token-generation paradigm struggles with pixel-level dense prediction. Existing RES methods either couple MLLMs with the parameter-heavy Segment Anything Model (SAM) with 632M network parameters or adopt SAM-free lightweight pipelines that sacrifice accuracy. To address the trade-off between performance and cost, we specifically propose MLLMSeg, a novel framework that fully exploits the inherent visual detail features encoded in the MLLM vision encoder without introducing an extra visual encoder. Besides, we propose a detail-enhanced and semantic-consistent feature fusion module (DSFF) that fully integrates the detail-related visual feature with the semantic-related feature output by the large language model (LLM) of MLLM. Finally, we establish a light-weight mask decoder with only 34M network parameters that optimally leverages detailed spatial features from the visual encoder and semantic features from the LLM to achieve precise mask prediction. Extensive experiments demonstrate that our method generally surpasses both SAM-based and SAM-free competitors, striking a better balance between performance and cost. Code is available at https://github.com/jcwang0602/MLLMSeg.

Hong Huang, Weixiang Sun, Zhijian Wu et al.

Recently, the rapid advancements of vision-language models, such as CLIP, leads to significant progress in zero-/few-shot anomaly detection (ZFSAD) tasks. However, most existing CLIP-based ZFSAD methods commonly assume prior knowledge of categories and rely on carefully crafted prompts tailored to specific scenarios. While such meticulously designed text prompts effectively capture semantic information in the textual space, they fall short of distinguishing normal and anomalous instances within the joint embedding space. Moreover, these ZFSAD methods are predominantly explored in industrial scenarios, with few efforts conducted to medical tasks. To this end, we propose an innovative framework for ZFSAD tasks in medical domain, denoted as IQE-CLIP. We reveal that query embeddings, which incorporate both textual and instance-aware visual information, are better indicators for abnormalities. Specifically, we first introduce class-based prompting tokens and learnable prompting tokens for better adaptation of CLIP to the medical domain. Then, we design an instance-aware query module (IQM) to extract region-level contextual information from both text prompts and visual features, enabling the generation of query embeddings that are more sensitive to anomalies. Extensive experiments conducted on six medical datasets demonstrate that IQE-CLIP achieves state-of-the-art performance on both zero-shot and few-shot tasks. We release our code and data at https://github.com/hongh0/IQE-CLIP/.

Yuhang Wang, Jun Li, Zhijian Wu et al.

Within the family of convolutional neural networks, InceptionNeXt has shown excellent competitiveness in image classification and a number of downstream tasks. Built on parallel one-dimensional strip convolutions, however, it suffers from limited ability of capturing spatial dependencies along different dimensions and fails to fully explore spatial modeling in local neighborhood. Besides, inherent locality constraints of convolution operations are detrimental to effective global context modeling. To overcome these limitations, we propose a novel backbone architecture termed InceptionMamba in this study. More specifically, the traditional one-dimensional strip convolutions are replaced by orthogonal band convolutions in our InceptionMamba to achieve cohesive spatial modeling. Furthermore, global contextual modeling can be achieved via a bottleneck Mamba module, facilitating enhanced cross-channel information fusion and enlarged receptive field. Extensive evaluations on classification and various downstream tasks demonstrate that the proposed InceptionMamba achieves state-of-the-art performance with superior parameter and computational efficiency. The source code will be available at https://github.com/Wake1021/InceptionMamba.

Zhengxu He, Jun Li, Zhijian Wu

Large-scale Vision-Language Models (VLMs) encode rich multimodal semantics that are highly beneficial for fine-grained visual categorization (FGVC). However, their prohibitive computational cost hinders practical deployment in resource-constrained environments. Although knowledge distillation contributes to transferring VLMs capacity to lightweight classifiers, conventional distillation mechanisms, which directly transfer from a generic VLM to a compact student, often yield suboptimal results due to severe architectural misalignment and introducing task-irrelevant information. To alleviate this limitation, we propose Distillation with Adaptive Intermediate Teacher transfer (DAIT) in this study, facilitating adaptive knowledge transfer from VLMs to lightweight students. DAIT introduces a trainable intermediate teacher that learns to transfer frozen VLMs representations under explicit supervision from the target fine-grained task. This intermediate teacher adaptively enhances discriminative visual cues, thereby producing compact and task-aligned knowledge that can be reliably distilled into lightweight models. Extensive evaluations on multiple FGVC benchmarks with diverse student architectures demonstrate that our method achieves respective performance gains of 12.63% and 8.34% on FGVC-Aircraft and CUB-200-2011 datasets, establishing DAIT as a principled paradigm for transferring from general-purpose VLMS to deployable fine-grained recognition models.

Lixin Chen, Chaomeng Chen, Jiale Zhou et al.

Despite the rapid progress of deep learning in video action recognition (VAR) in recent years, privacy leakage in videos remains a critical concern. Current state-of-the-art privacy-preserving methods often rely on anonymization. These methods suffer from (1) low concealment, where producing visually distorted videos that attract attackers' attention during transmission, and (2) spatiotemporal disruption, where degrading essential spatiotemporal features for accurate VAR. To address these issues, we propose StegaVAR, a novel framework that embeds action videos into ordinary cover videos and directly performs VAR in the steganographic domain for the first time. Throughout both data transmission and action analysis, the spatiotemporal information of hidden secret video remains complete, while the natural appearance of cover videos ensures the concealment of transmission. Considering the difficulty of steganographic domain analysis, we propose Secret Spatio-Temporal Promotion (STeP) and Cross-Band Difference Attention (CroDA) for analysis within the steganographic domain. STeP uses the secret video to guide spatiotemporal feature extraction in the steganographic domain during training. CroDA suppresses cover interference by capturing cross-band semantic differences. Experiments demonstrate that StegaVAR achieves superior VAR and privacy-preserving performance on widely used datasets. Moreover, our framework is effective for multiple steganographic models.

Mingyang Yu, Zhijian Wu, Dingjiang Huang

Hyperspectral image (HSI) reconstruction aims to recover 3D HSI from its degraded 2D measurements. Recently great progress has been made in deep learning-based methods, however, these methods often struggle to accurately capture high-frequency details of the HSI. To address this issue, this paper proposes a Spectral Diffusion Prior (SDP) that is implicitly learned from hyperspectral images using a diffusion model. Leveraging the powerful ability of the diffusion model to reconstruct details, this learned prior can significantly improve the performance when injected into the HSI model. To further improve the effectiveness of the learned prior, we also propose the Spectral Prior Injector Module (SPIM) to dynamically guide the model to recover the HSI details. We evaluate our method on two representative HSI methods: MST and BISRNet. Experimental results show that our method outperforms existing networks by about 0.5 dB, effectively improving the performance of HSI reconstruction.

Wenhui Liu, Zhijian Wu, Jingchao Wang et al.

Offline reinforcement learning seeks to derive improved policies entirely from historical data but often struggles with over-optimistic value estimates for out-of-distribution (OOD) actions. This issue is typically mitigated via policy constraint or conservative value regularization methods. However, these approaches may impose overly constraints or biased value estimates, potentially limiting performance improvements. To balance exploitation and restriction, we propose an Imagination-Limited Q-learning (ILQ) method, which aims to maintain the optimism that OOD actions deserve within appropriate limits. Specifically, we utilize the dynamics model to imagine OOD action-values, and then clip the imagined values with the maximum behavior values. Such design maintains reasonable evaluation of OOD actions to the furthest extent, while avoiding its over-optimism. Theoretically, we prove the convergence of the proposed ILQ under tabular Markov decision processes. Particularly, we demonstrate that the error bound between estimated values and optimality values of OOD state-actions possesses the same magnitude as that of in-distribution ones, thereby indicating that the bias in value estimates is effectively mitigated. Empirically, our method achieves state-of-the-art performance on a wide range of tasks in the D4RL benchmark.