Lihua Zhou

Lihua Zhou, Mao Ye, Xiatian Zhu et al.

Open-vocabulary object detection with vision-language models (VLMs) such as Grounding DINO suffers from performance degradation under test-time distribution shifts, primarily due to semantic misalignment between text embeddings and shifted visual embeddings of region proposals. While recent test-time adaptive object detection methods for VLM-based either rely on costly backpropagation or bypass semantic misalignment via external memory, none directly and efficiently align text and vision in a training-free manner. To address this, we propose Reward-Guided Semantic Evolution (RGSE), a training-free framework that directly refines the text embeddings at test time. Inspired by evolutionary search, RGSE treats text embedding adaptation as a semantic search process: it perturbs text embeddings as candidate variants, evaluates them via cosine similarity with current and historical high-confidence visual proposals as a reward signal, and fuses them into a refined embedding through reward-weighted averaging. Without any backpropagation, RGSE achieves state-of-the-art performance across multiple detection benchmarks while adding minimal computational overhead. Our code will be open source upon publication.

Jinlin Wu, Felix Holm, Chuxi Chen et al.

While foundation models have advanced surgical video analysis, current approaches rely predominantly on pixel-level reconstruction objectives that waste model capacity on low-level visual details - such as smoke, specular reflections, and fluid motion - rather than semantic structures essential for surgical understanding. We present UniSurg, a video-native foundation model that shifts the learning paradigm from pixel-level reconstruction to latent motion prediction. Built on the Video Joint Embedding Predictive Architecture (V-JEPA), UniSurg introduces three key technical innovations tailored to surgical videos: 1) motion-guided latent prediction to prioritize semantically meaningful regions, 2) spatiotemporal affinity self-distillation to enforce relational consistency, and 3) feature diversity regularization to prevent representation collapse in texture-sparse surgical scenes. To enable large-scale pretraining, we curate UniSurg-15M, the largest surgical video dataset to date, comprising 3,658 hours of video from 50 sources across 13 anatomical regions. Extensive experiments across 17 benchmarks demonstrate that UniSurg significantly outperforms state-of-the-art methods on surgical workflow recognition (+14.6% F1 on EgoSurgery, +10.3% on PitVis), action triplet recognition (39.54% mAP-IVT on CholecT50), skill assessment, polyp segmentation, and depth estimation. These results establish UniSurg as a new standard for universal, motion-oriented surgical video understanding.

Kaixiang Zhao, Mao Ye, Lihua Zhou et al.

Open-vocabulary object detection often fails under distribution shifts, as it can be misled by spurious correlations between non-causal visual attributes (e.g., brightness, texture) and object categories. Existing test-time adaptation (TTA) methods either depend on costly online optimization or perform global calibration, overlooking the attribute-specific nature of these failures. To address this, we propose FACTOR (counterFACtual training-free Test-time adaptation for Open-vocabulaRy object detection), a lightweight framework grounded in counterfactual reasoning. By perturbing test images along non-causal attributes and comparing region-level predictions between original and counterfactual views, FACTOR quantifies attribute sensitivity, semantic relevance, and prediction variation to selectively suppress attribute-dependent predictions-without parameter updates. Experiments on PASCAL-C, COCO-C, and FoggyCityscapes show that FACTOR consistently outperforms prior TTA methods, demonstrating that explicit counterfactual reasoning effectively improves robustness under distribution shifts.

Lihua Zhou, Mao Ye, Shuaifeng Li et al.

Test-time adaptation with pre-trained vision-language models, such as CLIP, aims to adapt the model to new, potentially out-of-distribution test data. Existing methods calculate the similarity between visual embedding and learnable class embeddings, which are initialized by text embeddings, for zero-shot image classification. In this work, we first analyze this process based on Bayes theorem, and observe that the core factors influencing the final prediction are the likelihood and the prior. However, existing methods essentially focus on adapting class embeddings to adapt likelihood, but they often ignore the importance of prior. To address this gap, we propose a novel approach, \textbf{B}ayesian \textbf{C}lass \textbf{A}daptation (BCA), which in addition to continuously updating class embeddings to adapt likelihood, also uses the posterior of incoming samples to continuously update the prior for each class embedding. This dual updating mechanism allows the model to better adapt to distribution shifts and achieve higher prediction accuracy. Our method not only surpasses existing approaches in terms of performance metrics but also maintains superior inference rates and memory usage, making it highly efficient and practical for real-world applications.

Lihua Zhou, Mao Ye, Nianxin Li et al.

Deep learning often struggles when training and test data distributions differ. Traditional domain generalization (DG) tackles this by including data from multiple source domains, which is impractical due to expensive data collection and annotation. Recent vision-language models like CLIP enable source-free domain generalization (SFDG) by using text prompts to simulate visual representations, reducing data demands. However, existing SFDG methods struggle with domain-specific confounders, limiting their generalization capabilities. To address this issue, we propose TDCRL (\textbf{T}ext-\textbf{D}riven \textbf{C}ausal \textbf{R}epresentation \textbf{L}earning), the first method to integrate causal inference into the SFDG setting. TDCRL operates in two steps: first, it employs data augmentation to generate style word vectors, combining them with class information to generate text embeddings to simulate visual representations; second, it trains a causal intervention network with a confounder dictionary to extract domain-invariant features. Grounded in causal learning, our approach offers a clear and effective mechanism to achieve robust, domain-invariant features, ensuring robust generalization. Extensive experiments on PACS, VLCS, OfficeHome, and DomainNet show state-of-the-art performance, proving TDCRL effectiveness in SFDG.

Lihua Zhou, Mao Ye, Shuaifeng Li et al.

Vision-language models (VLMs) such as CLIP and Grounding DINO have achieved remarkable success in object recognition and detection. However, their performance often degrades under real-world distribution shifts. Test-time adaptation (TTA) aims to mitigate this issue by adapting models during inference. Existing methods either rely on computationally expensive backpropagation, which hinders real-time deployment, or focus solely on likelihood adaptation, which overlooks the critical role of the prior. Our prior work, Bayesian Class Adaptation (BCA), addressed these shortcomings for object recognition by introducing a training-free framework that incorporates adaptive priors. Building upon this foundation, we now present Bayesian Class Adaptation plus (BCA+), a unified, training-free framework for TTA for both object recognition and detection. BCA+ introduces a dynamic cache that adaptively stores and updates class embeddings, spatial scales (for detection), and, crucially, adaptive class priors derived from historical predictions. We formulate adaptation as a Bayesian inference problem, where final predictions are generated by fusing the initial VLM output with a cache-based prediction. This cache-based prediction combines a dynamically updated likelihood (measuring feature and scale similarity) and a prior (reflecting the evolving class distribution). This dual-adaptation mechanism, coupled with uncertainty-guided fusion, enables BCA+ to correct both the model's semantic understanding and its contextual confidence. As a training-free method requiring no backpropagation, BCA+ is highly efficient. Extensive experiments demonstrate that BCA+ achieves state-of-the-art performance on both recognition and detection benchmarks.

Xusheng Liang, Lihua Zhou, Nianxin Li et al.

Vision-Language Models (VLMs), such as CLIP, have demonstrated remarkable zero-shot capabilities in various computer vision tasks. However, their application to medical imaging remains challenging due to the high variability and complexity of medical data. Specifically, medical images often exhibit significant domain shifts caused by various confounders, including equipment differences, procedure artifacts, and imaging modes, which can lead to poor generalization when models are applied to unseen domains. To address this limitation, we propose Multimodal Causal-Driven Representation Learning (MCDRL), a novel framework that integrates causal inference with the VLM to tackle domain generalization in medical image segmentation. MCDRL is implemented in two steps: first, it leverages CLIP's cross-modal capabilities to identify candidate lesion regions and construct a confounder dictionary through text prompts, specifically designed to represent domain-specific variations; second, it trains a causal intervention network that utilizes this dictionary to identify and eliminate the influence of these domain-specific variations while preserving the anatomical structural information critical for segmentation tasks. Extensive experiments demonstrate that MCDRL consistently outperforms competing methods, yielding superior segmentation accuracy and exhibiting robust generalizability.

Lihua Zhou, Mao Ye, Xinpeng Li et al.

Recent works in domain adaptation always learn domain invariant features to mitigate the gap between the source and target domains by adversarial methods. The category information are not sufficiently used which causes the learned domain invariant features are not enough discriminative. We propose a new domain adaptation method based on prototype construction which likes capturing data cluster centers. Specifically, it consists of two parts: disentanglement and reconstruction. First, the domain specific features and domain invariant features are disentangled from the original features. At the same time, the domain prototypes and class prototypes of both domains are estimated. Then, a reconstructor is trained by reconstructing the original features from the disentangled domain invariant features and domain specific features. By this reconstructor, we can construct prototypes for the original features using class prototypes and domain prototypes correspondingly. In the end, the feature extraction network is forced to extract features close to these prototypes. Our contribution lies in the technical use of the reconstructor to obtain the original feature prototypes which helps to learn compact and discriminant features. As far as we know, this idea is proposed for the first time. Experiment results on several public datasets confirm the state-of-the-art performance of our method.