Yuxiao Wang

Yunhao Ge, Jie Ren, Andrew Gallagher et al. · deepmind

Multi-modal image-text models such as CLIP and LiT have demonstrated impressive performance on image classification benchmarks and their zero-shot generalization ability is particularly exciting. While the top-5 zero-shot accuracies of these models are very high, the top-1 accuracies are much lower (over 25% gap in some cases). We investigate the reasons for this performance gap and find that many of the failure cases are caused by ambiguity in the text prompts. First, we develop a simple and efficient zero-shot post-hoc method to identify images whose top-1 prediction is likely to be incorrect, by measuring consistency of the predictions w.r.t. multiple prompts and image transformations. We show that our procedure better predicts mistakes, outperforming the popular max logit baseline on selective prediction tasks. Next, we propose a simple and efficient way to improve accuracy on such uncertain images by making use of the WordNet hierarchy; specifically we augment the original class by incorporating its parent and children from the semantic label hierarchy, and plug the augmentation into text prompts. We conduct experiments on both CLIP and LiT models with five different ImageNet-based datasets. For CLIP, our method improves the top-1 accuracy by 17.13% on the uncertain subset and 3.6% on the entire ImageNet validation set. We also show that our method improves across ImageNet shifted datasets, four other datasets, and other model architectures such as LiT. The proposed method is hyperparameter-free, requires no additional model training and can be easily scaled to other large multi-modal architectures. Code is available at https://github.com/gyhandy/Hierarchy-CLIP.

Zheng Xu, Maxwell Collins, Yuxiao Wang et al.

Small on-device models have been successfully trained with user-level differential privacy (DP) for next word prediction and image classification tasks in the past. However, existing methods can fail when directly applied to learn embedding models using supervised training data with a large class space. To achieve user-level DP for large image-to-embedding feature extractors, we propose DP-FedEmb, a variant of federated learning algorithms with per-user sensitivity control and noise addition, to train from user-partitioned data centralized in the datacenter. DP-FedEmb combines virtual clients, partial aggregation, private local fine-tuning, and public pretraining to achieve strong privacy utility trade-offs. We apply DP-FedEmb to train image embedding models for faces, landmarks and natural species, and demonstrate its superior utility under same privacy budget on benchmark datasets DigiFace, EMNIST, GLD and iNaturalist. We further illustrate it is possible to achieve strong user-level DP guarantees of $ε<4$ while controlling the utility drop within 5%, when millions of users can participate in training.

Yuxiao Wang, Yu Lei, Li Cui et al.

Human-object interaction (HOI) detection plays a key role in high-level visual understanding, facilitating a deep comprehension of human activities. Specifically, HOI detection aims to locate the humans and objects involved in interactions within images or videos and classify the specific interactions between them. The success of this task is influenced by several key factors, including the accurate localization of human and object instances, as well as the correct classification of object categories and interaction relationships. This paper systematically summarizes and discusses the recent work in image-based HOI detection. First, the mainstream datasets involved in HOI relationship detection are introduced. Furthermore, starting with two-stage methods and end-to-end one-stage detection approaches, this paper comprehensively discusses the current developments in image-based HOI detection, analyzing the strengths and weaknesses of these two methods. Additionally, the advancements of zero-shot learning, weakly supervised learning, and the application of large-scale language models in HOI detection are discussed. Finally, the current challenges in HOI detection are outlined, and potential research directions and future trends are explored.

Imagen-Team-Google, Jason Baldridge, Jakob Bauer et al.

We introduce Imagen 3, a latent diffusion model that generates high quality images from text prompts. We describe our quality and responsibility evaluations. Imagen 3 is preferred over other state-of-the-art (SOTA) models at the time of evaluation. In addition, we discuss issues around safety and representation, as well as methods we used to minimize the potential harm of our models.

Yuxiao Wang, Yu Lei, Zhenao Wei et al.

The task of Human-Object conTact (HOT) detection involves identifying the specific areas of the human body that are touching objects. Nevertheless, current models are restricted to just one type of image, often leading to too much segmentation in areas with little interaction, and struggling to maintain category consistency within specific regions. To tackle this issue, a HOT framework, termed \textbf{P3HOT}, is proposed, which blends \textbf{P}rompt guidance and human \textbf{P}roximal \textbf{P}erception. To begin with, we utilize a semantic-driven prompt mechanism to direct the network's attention towards the relevant regions based on the correlation between image and text. Then a human proximal perception mechanism is employed to dynamically perceive key depth range around the human, using learnable parameters to effectively eliminate regions where interactions are not expected. Calculating depth resolves the uncertainty of the overlap between humans and objects in a 2D perspective, providing a quasi-3D viewpoint. Moreover, a Regional Joint Loss (RJLoss) has been created as a new loss to inhibit abnormal categories in the same area. A new evaluation metric called ``AD-Acc.'' is introduced to address the shortcomings of existing methods in addressing negative samples. Comprehensive experimental results demonstrate that our approach achieves state-of-the-art performance in four metrics across two benchmark datasets. Specifically, our model achieves an improvement of \textbf{0.7}$\uparrow$, \textbf{2.0}$\uparrow$, \textbf{1.6}$\uparrow$, and \textbf{11.0}$\uparrow$ in SC-Acc., mIoU, wIoU, and AD-Acc. metrics, respectively, on the HOT-Annotated dataset. The sources code are available at https://github.com/YuxiaoWang-AI/P3HOT.

Long Zhao, Yuxiao Wang, Jiaping Zhao et al.

We introduce a novel representation learning method to disentangle pose-dependent as well as view-dependent factors from 2D human poses. The method trains a network using cross-view mutual information maximization (CV-MIM) which maximizes mutual information of the same pose performed from different viewpoints in a contrastive learning manner. We further propose two regularization terms to ensure disentanglement and smoothness of the learned representations. The resulting pose representations can be used for cross-view action recognition. To evaluate the power of the learned representations, in addition to the conventional fully-supervised action recognition settings, we introduce a novel task called single-shot cross-view action recognition. This task trains models with actions from only one single viewpoint while models are evaluated on poses captured from all possible viewpoints. We evaluate the learned representations on standard benchmarks for action recognition, and show that (i) CV-MIM performs competitively compared with the state-of-the-art models in the fully-supervised scenarios; (ii) CV-MIM outperforms other competing methods by a large margin in the single-shot cross-view setting; (iii) and the learned representations can significantly boost the performance when reducing the amount of supervised training data. Our code is made publicly available at https://github.com/google-research/google-research/tree/master/poem

Yuxiao Wang, Di Wu, Feng Liu et al.

This paper proposes a composable fine-tuning method that integrates graph structural priors with modular adapters to address the high computational cost and structural instability faced by large-scale pre-trained models in multi-task adaptation. The method introduces a relation matrix to model dependencies among tasks, explicitly encoding correlations between nodes and paths into graph structural priors, which provide unified structural constraints for adapter weight allocation and path selection. Modular adapters are embedded into different layers through low-rank mapping and a pluggable mechanism, enabling efficient cross-task composition and reuse under prior guidance. This mechanism not only improves parameter efficiency and training stability but also alleviates path conflicts and redundant computation in multi-task scenarios. Furthermore, experiments on hyperparameter sensitivity, environmental sensitivity, and data sensitivity are conducted to systematically analyze key factors such as routing temperature, gating thresholds, and relation matrix regularization strength, verifying the consistency and superior performance of the method under structural constraints. The results demonstrate that the proposed framework significantly enhances task prediction accuracy, adapter weight allocation precision, and overall computational efficiency while maintaining model lightweight design, highlighting the synergistic advantages of graph priors and modular mechanisms in composable fine-tuning.

Yihan Xue, Yuxiao Wang, Ao Zhu et al.

Addressing the diverse fault morphologies, complex dependencies, and time-varying operational states in microservice distributed systems, this paper proposes a distributed fault discrimination model based on temporal graph neural networks. This model characterizes the microservice operation process as a dynamic graph sequence evolving, and performs joint representation learning of temporal modeling and structural interactions within a unified framework. First, service-level multi-source observation signals are aligned and characterized to construct node feature sequences and their corresponding time-dependent dependencies. Then, a temporal coding module is introduced to extract the dynamic evolution representation of service states, and at each time step, attention-based structured message passing is used to characterize dependency interactions and propagation associations, forming a structure-enhanced temporal node representation. Furthermore, a dual readout mechanism is employed to aggregate the node and temporal dimensions, obtaining a system-level global representation and outputting the fault category distribution. Finally, supervised learning objectives are used to optimize model parameters, enabling the model to learn stable discrimination evidence under complex interactions and multi-source noise conditions. Comparative experimental results show that the proposed method achieves superior performance on multiple evaluation metrics, validating the effectiveness of jointly modeling temporal evolution and dependency structures in improving the distributed fault discrimination capability of microservices.

Yuxiao Wang, Wenpeng Neng, Zhenao Wei et al.

Human-object contact (HOT) is designed to accurately identify the areas where humans and objects come into contact. Current methods frequently fail to account for scenarios where objects are frequently blocking the view, resulting in inaccurate identification of contact areas. To tackle this problem, we suggest using a perspective interaction HOT detector called PIHOT, which utilizes a depth map generation model to offer depth information of humans and objects related to the camera, thereby preventing false interaction detection. Furthermore, we use mask dilatation and object restoration techniques to restore the texture details in covered areas, improve the boundaries between objects, and enhance the perception of humans interacting with objects. Moreover, a spatial awareness perception is intended to concentrate on the characteristic features close to the points of contact. The experimental results show that the PIHOT algorithm achieves state-of-the-art performance on three benchmark datasets for HOT detection tasks. Compared to the most recent DHOT, our method enjoys an average improvement of 13%, 27.5%, 16%, and 18.5% on SC-Acc., C-Acc., mIoU, and wIoU metrics, respectively.

Weiying Xue, Qi Liu, Qiwei Xiong et al.

Human-object interaction (HOI) detection aims to locate human-object pairs and identify their interaction categories in images. Most existing methods primarily focus on supervised learning, which relies on extensive manual HOI annotations. In this paper, we propose a novel framework, termed Knowledge Integration to HOI (KI2HOI), that effectively integrates the knowledge of visual-language model to improve zero-shot HOI detection. Specifically, the verb feature learning module is designed based on visual semantics, by employing the verb extraction decoder to convert corresponding verb queries into interaction-specific category representations. We develop an effective additive self-attention mechanism to generate more comprehensive visual representations. Moreover, the innovative interaction representation decoder effectively extracts informative regions by integrating spatial and visual feature information through a cross-attention mechanism. To deal with zero-shot learning in low-data, we leverage a priori knowledge from the CLIP text encoder to initialize the linear classifier for enhanced interaction understanding. Extensive experiments conducted on the mainstream HICO-DET and V-COCO datasets demonstrate that our model outperforms the previous methods in various zero-shot and full-supervised settings.

Xiao Wang, Ke Tang, Xingyuan Dai et al.

In public roads, autonomous vehicles (AVs) face the challenge of frequent interactions with human-driven vehicles (HDVs), which render uncertain driving behavior due to varying social characteristics among humans. To effectively assess the risks prevailing in the vicinity of AVs in social interactive traffic scenarios and achieve safe autonomous driving, this article proposes a social-suitable and safety-sensitive trajectory planning (S4TP) framework. Specifically, S4TP integrates the Social-Aware Trajectory Prediction (SATP) and Social-Aware Driving Risk Field (SADRF) modules. SATP utilizes Transformers to effectively encode the driving scene and incorporates an AV's planned trajectory during the prediction decoding process. SADRF assesses the expected surrounding risk degrees during AVs-HDVs interactions, each with different social characteristics, visualized as two-dimensional heat maps centered on the AV. SADRF models the driving intentions of the surrounding HDVs and predicts trajectories based on the representation of vehicular interactions. S4TP employs an optimization-based approach for motion planning, utilizing the predicted HDVs'trajectories as input. With the integration of SADRF, S4TP executes real-time online optimization of the planned trajectory of AV within lowrisk regions, thus improving the safety and the interpretability of the planned trajectory. We have conducted comprehensive tests of the proposed method using the SMARTS simulator. Experimental results in complex social scenarios, such as unprotected left turn intersections, merging, cruising, and overtaking, validate the superiority of our proposed S4TP in terms of safety and rationality. S4TP achieves a pass rate of 100% across all scenarios, surpassing the current state-of-the-art methods Fanta of 98.25% and Predictive-Decision of 94.75%.

Shaoheng Fang, Rui Ye, Wenhao Wang et al.

Roadside unit (RSU) can significantly improve the safety and robustness of autonomous vehicles through Vehicle-to-Everything (V2X) communication. Currently, the usage of a single RSU mainly focuses on real-time inference and V2X collaboration, while neglecting the potential value of the high-quality data collected by RSU sensors. Integrating the vast amounts of data from numerous RSUs can provide a rich source of data for model training. However, the absence of ground truth annotations and the difficulty of transmitting enormous volumes of data are two inevitable barriers to fully exploiting this hidden value. In this paper, we introduce FedRSU, an innovative federated learning framework for self-supervised scene flow estimation. In FedRSU, we present a recurrent self-supervision training paradigm, where for each RSU, the scene flow prediction of points at every timestamp can be supervised by its subsequent future multi-modality observation. Another key component of FedRSU is federated learning, where multiple devices collaboratively train an ML model while keeping the training data local and private. With the power of the recurrent self-supervised learning paradigm, FL is able to leverage innumerable underutilized data from RSU. To verify the FedRSU framework, we construct a large-scale multi-modality dataset RSU-SF. The dataset consists of 17 RSU clients, covering various scenarios, modalities, and sensor settings. Based on RSU-SF, we show that FedRSU can greatly improve model performance in ITS and provide a comprehensive benchmark under diverse FL scenarios. To the best of our knowledge, we provide the first real-world LiDAR-camera multi-modal dataset and benchmark for the FL community.

Ben Wan, Tianyi Zheng, Zhaoyu Chen et al.

Diffusion Models (DMs) have impressive capabilities among generation models, but are limited to slower inference speeds and higher computational costs. Previous works utilize one-shot structure pruning to derive lightweight DMs from pre-trained ones, but this approach often leads to a significant drop in generation quality and may result in the removal of crucial weights. Thus we propose a iterative pruning method based on gradient flow, including the gradient flow pruning process and the gradient flow pruning criterion. We employ a progressive soft pruning strategy to maintain the continuity of the mask matrix and guide it along the gradient flow of the energy function based on the pruning criterion in sparse space, thereby avoiding the sudden information loss typically caused by one-shot pruning. Gradient-flow based criterion prune parameters whose removal increases the gradient norm of loss function and can enable fast convergence for a pruned model in iterative pruning stage. Our extensive experiments on widely used datasets demonstrate that our method achieves superior performance in efficiency and consistency with pre-trained models.

Qi Liu, Weiying Xue, Yuxiao Wang et al.

The video visual relation detection (VidVRD) task is to identify objects and their relationships in videos, which is challenging due to the dynamic content, high annotation costs, and long-tailed distribution of relations. Visual language models (VLMs) help explore open-vocabulary visual relation detection tasks, yet often overlook the connections between various visual regions and their relations. Moreover, using VLMs to directly identify visual relations in videos poses significant challenges because of the large disparity between images and videos. Therefore, we propose a novel open-vocabulary VidVRD framework, termed OpenVidVRD, which transfers VLMs' rich knowledge and powerful capabilities to improve VidVRD tasks through prompt learning. Specificall y, We use VLM to extract text representations from automatically generated region captions based on the video's regions. Next, we develop a spatiotemporal refiner module to derive object-level relationship representations in the video by integrating cross-modal spatiotemporal complementary information. Furthermore, a prompt-driven strategy to align semantic spaces is employed to harness the semantic understanding of VLMs, enhancing the overall generalization ability of OpenVidVRD. Extensive experiments conducted on the VidVRD and VidOR public datasets show that the proposed model outperforms existing methods.

Zhimin Qiu, Di Wu, Feng Liu et al.

This paper proposes a structure-aware decoding method based on large language models to address the difficulty of traditional approaches in maintaining both semantic integrity and structural consistency in nested and overlapping entity extraction tasks. The method introduces a candidate span generation mechanism and structured attention modeling to achieve unified modeling of entity boundaries, hierarchical relationships, and cross-dependencies. The model first uses a pretrained language model to obtain context-aware semantic representations, then captures multi-granular entity span features through candidate representation combinations, and introduces hierarchical structural constraints during decoding to ensure consistency between semantics and structure. To enhance stability in complex scenarios, the model jointly optimizes classification loss and structural consistency loss, maintaining high recognition accuracy under multi-entity co-occurrence and long-sentence dependency conditions. Experiments conducted on the ACE 2005 dataset demonstrate significant improvements in Accuracy, Precision, Recall, and F1-Score, particularly in nested and overlapping entity recognition, where the model shows stronger boundary localization and structural modeling capability. This study verifies the effectiveness of structure-aware decoding in complex semantic extraction tasks, provides a new perspective for developing language models with hierarchical understanding, and establishes a methodological foundation for high-precision information extraction.

Nan Zhuang, Wenshuo Wang, Lekai Qian et al.

Recent studies have demonstrated that some Large Language Models exhibit choice-supportive bias (CSB) when performing evaluations, systematically favoring their chosen options and potentially compromising the objectivity of AI-assisted decision making. While existing debiasing approaches primarily target demographic and social biases, methods for addressing cognitive biases in LLMs remain largely unexplored. In this work, we present the first solution to address CSB through Reasoning Dependency Generation (RDG), a novel framework for generating unbiased reasoning data to mitigate choice-supportive bias through fine-tuning. RDG automatically constructs balanced reasoning QA pairs, explicitly (un)modeling the dependencies between choices, evidences, and justifications. Our approach is able to generate a large-scale dataset of QA pairs across domains, incorporating Contextual Dependency Data and Dependency Decouple Data. Experiments show that LLMs fine-tuned on RDG-generated data demonstrate a 81.5% improvement in memory-based experiments and 94.3% improvement in the evaluation-based experiment, while maintaining similar performance on standard BBQ benchmarks. This work pioneers an approach for addressing cognitive biases in LLMs and contributes to the development of more reliable AI-assisted decision support systems.

Wenhao Wang, Longqi Cai, Taihong Xiao et al.

This paper presents a systematic study of scaling laws for the deepfake detection task. Specifically, we analyze the model performance against the number of real image domains, deepfake generation methods, and training images. Since no existing dataset meets the scale requirements for this research, we construct ScaleDF, the largest dataset to date in this field, which contains over 5.8 million real images from 51 different datasets (domains) and more than 8.8 million fake images generated by 102 deepfake methods. Using ScaleDF, we observe power-law scaling similar to that shown in large language models (LLMs). Specifically, the average detection error follows a predictable power-law decay as either the number of real domains or the number of deepfake methods increases. This key observation not only allows us to forecast the number of additional real domains or deepfake methods required to reach a target performance, but also inspires us to counter the evolving deepfake technology in a data-centric manner. Beyond this, we examine the role of pre-training and data augmentations in deepfake detection under scaling, as well as the limitations of scaling itself.

Yuxiao Wang, Yu Lei, Wolin Liang et al.

People control their bodies to establish contact with the environment. To comprehensively understand actions across diverse visual contexts, it is essential to simultaneously consider \textbf{what} action is occurring and \textbf{where} it is happening. Current methodologies, however, often inadequately capture this duality, typically failing to jointly model both action semantics and their spatial contextualization within scenes. To bridge this gap, we introduce a novel vision task that simultaneously predicts high-level action semantics and fine-grained body-part contact regions. Our proposed framework, PaIR-Net, comprises three key components: the Contact Prior Aware Module (CPAM) for identifying contact-relevant body parts, the Prior-Guided Concat Segmenter (PGCS) for pixel-wise contact segmentation, and the Interaction Inference Module (IIM) responsible for integrating global interaction relationships. To facilitate this task, we present PaIR (Part-aware Interaction Representation), a comprehensive dataset containing 13,979 images that encompass 654 actions, 80 object categories, and 17 body parts. Experimental evaluation demonstrates that PaIR-Net significantly outperforms baseline approaches, while ablation studies confirm the efficacy of each architectural component. The code and dataset will be released upon publication.

Yuxiao Wang, Wolin Liang, Yu Lei et al.

Human-Object Interaction (HOI) detection aims to localize human-object pairs and recognize their interactions in images. Although DETR-based methods have recently emerged as the mainstream framework for HOI detection, they still suffer from a key limitation: Randomly initialized queries lack explicit semantics, leading to suboptimal detection performance. To address this challenge, we propose QueryCraft, a novel plug-and-play HOI detection framework that incorporates semantic priors and guided feature learning through transformer-based query initialization. Central to our approach is \textbf{ACTOR} (\textbf{A}ction-aware \textbf{C}ross-modal \textbf{T}ransf\textbf{OR}mer), a cross-modal Transformer encoder that jointly attends to visual regions and textual prompts to extract action-relevant features. Rather than merely aligning modalities, ACTOR leverages language-guided attention to infer interaction semantics and produce semantically meaningful query representations. To further enhance object-level query quality, we introduce a \textbf{P}erceptual \textbf{D}istilled \textbf{Q}uery \textbf{D}ecoder (\textbf{PDQD}), which distills object category awareness from a pre-trained detector to serve as object query initiation. This dual-branch query initialization enables the model to generate more interpretable and effective queries for HOI detection. Extensive experiments on HICO-Det and V-COCO benchmarks demonstrate that our method achieves state-of-the-art performance and strong generalization. Code will be released upon publication.

Qi Liu, Yuxiao Wang, Xinyu Jiang et al.

Recent human-object interaction (HOI) detection methods depend on extensively annotated image datasets, which require a significant amount of manpower. In this paper, we propose a novel self-adaptive, language-driven HOI detection method, termed FreeA. This method leverages the adaptability of the text-image model to generate latent HOI labels without requiring manual annotation. Specifically, FreeA aligns image features of human-object pairs with HOI text templates and employs a knowledge-based masking technique to decrease improbable interactions. Furthermore, FreeA implements a proposed method for matching interaction correlations to increase the probability of actions associated with a particular action, thereby improving the generated HOI labels. Experiments on two benchmark datasets showcase that FreeA achieves state-of-the-art performance among weakly supervised HOI competitors. Our proposal gets +\textbf{13.29} (\textbf{159\%$\uparrow$}) mAP and +\textbf{17.30} (\textbf{98\%$\uparrow$}) mAP than the newest ``Weakly'' supervised model, and +\textbf{7.19} (\textbf{28\%$\uparrow$}) mAP and +\textbf{14.69} (\textbf{34\%$\uparrow$}) mAP than the latest ``Weakly+'' supervised model, respectively, on HICO-DET and V-COCO datasets, more accurate in localizing and classifying the interactive actions. The source code will be made public.

Ting Liu, Jennifer J. Sun, Long Zhao et al.

Recognition of human poses and actions is crucial for autonomous systems to interact smoothly with people. However, cameras generally capture human poses in 2D as images and videos, which can have significant appearance variations across viewpoints that make the recognition tasks challenging. To address this, we explore recognizing similarity in 3D human body poses from 2D information, which has not been well-studied in existing works. Here, we propose an approach to learning a compact view-invariant embedding space from 2D body joint keypoints, without explicitly predicting 3D poses. Input ambiguities of 2D poses from projection and occlusion are difficult to represent through a deterministic mapping, and therefore we adopt a probabilistic formulation for our embedding space. Experimental results show that our embedding model achieves higher accuracy when retrieving similar poses across different camera views, in comparison with 3D pose estimation models. We also show that by training a simple temporal embedding model, we achieve superior performance on pose sequence retrieval and largely reduce the embedding dimension from stacking frame-based embeddings for efficient large-scale retrieval. Furthermore, in order to enable our embeddings to work with partially visible input, we further investigate different keypoint occlusion augmentation strategies during training. We demonstrate that these occlusion augmentations significantly improve retrieval performance on partial 2D input poses. Results on action recognition and video alignment demonstrate that using our embeddings without any additional training achieves competitive performance relative to other models specifically trained for each task.