Search

Recently, large foundation models, including large language models (LLMs) and large vision-language models (LVLMs), have become essential tools in critical fields such as law, finance, and healthcare. As these models increasingly integrate into our daily life, it is necessary to conduct moral evaluation to ensure that their outputs align with human values and remain within moral boundaries. Previous works primarily focus on LLMs, proposing moral datasets and benchmarks limited to text modality. However, given the rapid development of LVLMs, there is still a lack of multimodal moral evaluation methods. To bridge this gap, we introduce M$^3$oralBench, the first MultiModal Moral Benchmark for LVLMs. M$^3$oralBench expands the everyday moral scenarios in Moral Foundations Vignettes (MFVs) and employs the text-to-image diffusion model, SD3.0, to create corresponding scenario images. It conducts moral evaluation across six moral foundations of Moral Foundations Theory (MFT) and encompasses tasks in moral judgement, moral classification, and moral response, providing a comprehensive assessment of model performance in multimodal moral understanding and reasoning. Extensive experiments on 10 popular open-source and closed-source LVLMs demonstrate that M$^3$oralBench is a challenging benchmark, exposing notable moral limitations in current models. Our benchmark is publicly available.

Large Vision-Language Models (LVLMs) exhibit impressive potential across various tasks but also face significant privacy risks, limiting their practical applications. Current researches on privacy assessment for LVLMs is limited in scope, with gaps in both assessment dimensions and privacy categories. To bridge this gap, we propose Multi-P$^2$A, a comprehensive benchmark for evaluating the privacy preservation capabilities of LVLMs in terms of privacy awareness and leakage. Privacy awareness measures the model's ability to recognize the privacy sensitivity of input data, while privacy leakage assesses the risk of the model unintentionally disclosing privacy information in its output. We design a range of sub-tasks to thoroughly evaluate the model's privacy protection offered by LVLMs. Multi-P$^2$A covers 26 categories of personal privacy, 15 categories of trade secrets, and 18 categories of state secrets, totaling 31,962 samples. Based on Multi-P$^2$A, we evaluate the privacy preservation capabilities of 21 open-source and 2 closed-source LVLMs. Our results reveal that current LVLMs generally pose a high risk of facilitating privacy breaches, with vulnerabilities varying across personal privacy, trade secret, and state secret.

Human-centric perceptions play a crucial role in real-world applications. While recent human-centric works have achieved impressive progress, these efforts are often constrained to the visual domain and lack interaction with human instructions, limiting their applicability in broader scenarios such as chatbots and sports analysis. This paper introduces Referring Human Perceptions, where a referring prompt specifies the person of interest in an image. To tackle the new task, we propose RefHCM (Referring Human-Centric Model), a unified framework to integrate a wide range of human-centric referring tasks. Specifically, RefHCM employs sequence mergers to convert raw multimodal data -- including images, text, coordinates, and parsing maps -- into semantic tokens. This standardized representation enables RefHCM to reformulate diverse human-centric referring tasks into a sequence-to-sequence paradigm, solved using a plain encoder-decoder transformer architecture. Benefiting from a unified learning strategy, RefHCM effectively facilitates knowledge transfer across tasks and exhibits unforeseen capabilities in handling complex reasoning. This work represents the first attempt to address referring human perceptions with a general-purpose framework, while simultaneously establishing a corresponding benchmark that sets new standards for the field. Extensive experiments showcase RefHCM's competitive and even superior performance across multiple human-centric referring tasks. The code and data are publicly at https://github.com/JJJYmmm/RefHCM., Comment: 13 pages

This paper presents a novel approach that enables autoregressive video generation with high efficiency. We propose to reformulate the video generation problem as a non-quantized autoregressive modeling of temporal frame-by-frame prediction and spatial set-by-set prediction. Unlike raster-scan prediction in prior autoregressive models or joint distribution modeling of fixed-length tokens in diffusion models, our approach maintains the causal property of GPT-style models for flexible in-context capabilities, while leveraging bidirectional modeling within individual frames for efficiency. With the proposed approach, we train a novel video autoregressive model without vector quantization, termed NOVA. Our results demonstrate that NOVA surpasses prior autoregressive video models in data efficiency, inference speed, visual fidelity, and video fluency, even with a much smaller model capacity, i.e., 0.6B parameters. NOVA also outperforms state-of-the-art image diffusion models in text-to-image generation tasks, with a significantly lower training cost. Additionally, NOVA generalizes well across extended video durations and enables diverse zero-shot applications in one unified model. Code and models are publicly available at https://github.com/baaivision/NOVA., Comment: 22 pages, 16 figures

Human pose plays a crucial role in the digital age. While recent works have achieved impressive progress in understanding and generating human poses, they often support only a single modality of control signals and operate in isolation, limiting their application in real-world scenarios. This paper presents UniPose, a framework employing Large Language Models (LLMs) to comprehend, generate, and edit human poses across various modalities, including images, text, and 3D SMPL poses. Specifically, we apply a pose tokenizer to convert 3D poses into discrete pose tokens, enabling seamless integration into the LLM within a unified vocabulary. To further enhance the fine-grained pose perception capabilities, we facilitate UniPose with a mixture of visual encoders, among them a pose-specific visual encoder. Benefiting from a unified learning strategy, UniPose effectively transfers knowledge across different pose-relevant tasks, adapts to unseen tasks, and exhibits extended capabilities. This work serves as the first attempt at building a general-purpose framework for pose comprehension, generation, and editing. Extensive experiments highlight UniPose's competitive and even superior performance across various pose-relevant tasks.

The primary goal of out-of-distribution (OOD) detection tasks is to identify inputs with semantic shifts, i.e., if samples from novel classes are absent in the in-distribution (ID) dataset used for training, we should reject these OOD samples rather than misclassifying them into existing ID classes. However, we find the current definition of "semantic shift" is ambiguous, which renders certain OOD testing protocols intractable for the post-hoc OOD detection methods based on a classifier trained on the ID dataset. In this paper, we offer a more precise definition of the Semantic Space and the Covariate Space for the ID distribution, allowing us to theoretically analyze which types of OOD distributions make the detection task intractable. To avoid the flaw in the existing OOD settings, we further define the "Tractable OOD" setting which ensures the distinguishability of OOD and ID distributions for the post-hoc OOD detection methods. Finally, we conduct several experiments to demonstrate the necessity of our definitions and validate the correctness of our theorems., Comment: v1

Pre-trained vision-language models (e.g., CLIP) have shown powerful zero-shot transfer capabilities. But they still struggle with domain shifts and typically require labeled data to adapt to downstream tasks, which could be costly. In this work, we aim to leverage unlabeled data that naturally spans multiple domains to enhance the transferability of vision-language models. Under this unsupervised multi-domain setting, we have identified inherent model bias within CLIP, notably in its visual and text encoders. Specifically, we observe that CLIP's visual encoder tends to prioritize encoding domain over discriminative category information, meanwhile its text encoder exhibits a preference for domain-relevant classes. To mitigate this model bias, we propose a training-free and label-free feature calibration method, Unsupervised Multi-domain Feature Calibration (UMFC). UMFC estimates image-level biases from domain-specific features and text-level biases from the direction of domain transition. These biases are subsequently subtracted from original image and text features separately, to render them domain-invariant. We evaluate our method on multiple settings including transductive learning and test-time adaptation. Extensive experiments show that our method outperforms CLIP and performs on par with the state-of-the-arts that need additional annotations or optimization. Our code is available at https://github.com/GIT-LJc/UMFC., Comment: NeurIPS 2024

Emotion Recognition (ER) is the process of identifying human emotions from given data. Currently, the field heavily relies on facial expression recognition (FER) because facial expressions contain rich emotional cues. However, it is important to note that facial expressions may not always precisely reflect genuine emotions and FER-based results may yield misleading ER. To understand and bridge this gap between FER and ER, we introduce eye behaviors as an important emotional cues for the creation of a new Eye-behavior-aided Multimodal Emotion Recognition (EMER) dataset. Different from existing multimodal ER datasets, the EMER dataset employs a stimulus material-induced spontaneous emotion generation method to integrate non-invasive eye behavior data, like eye movements and eye fixation maps, with facial videos, aiming to obtain natural and accurate human emotions. Notably, for the first time, we provide annotations for both ER and FER in the EMER, enabling a comprehensive analysis to better illustrate the gap between both tasks. Furthermore, we specifically design a new EMERT architecture to concurrently enhance performance in both ER and FER by efficiently identifying and bridging the emotion gap between the two.Specifically, our EMERT employs modality-adversarial feature decoupling and multi-task Transformer to augment the modeling of eye behaviors, thus providing an effective complement to facial expressions. In the experiment, we introduce seven multimodal benchmark protocols for a variety of comprehensive evaluations of the EMER dataset. The results show that the EMERT outperforms other state-of-the-art multimodal methods by a great margin, revealing the importance of modeling eye behaviors for robust ER. To sum up, we provide a comprehensive analysis of the importance of eye behaviors in ER, advancing the study on addressing the gap between FER and ER for more robust ER performance., Comment: The paper is part of ongoing work and we request to withdraw it from arXiv to revise it further. And The paper was submitted without agreement from all co-authors

Current Face Anti-spoofing (FAS) models tend to make overly confident predictions even when encountering unfamiliar scenarios or unknown presentation attacks, which leads to serious potential risks. To solve this problem, we propose a Confidence Aware Face Anti-spoofing (CA-FAS) model, which is aware of its capability boundary, thus achieving reliable liveness detection within this boundary. To enable the CA-FAS to "know what it doesn't know", we propose to estimate its confidence during the prediction of each sample. Specifically, we build Gaussian distributions for both the live faces and the known attacks. The prediction confidence for each sample is subsequently assessed using the Mahalanobis distance between the sample and the Gaussians for the "known data". We further introduce the Mahalanobis distance-based triplet mining to optimize the parameters of both the model and the constructed Gaussians as a whole. Extensive experiments show that the proposed CA-FAS can effectively recognize samples with low prediction confidence and thus achieve much more reliable performance than other FAS models by filtering out samples that are beyond its reliable range., Comment: v1

Although multimodal large language models (MLLMs) have achieved promising results on a wide range of vision-language tasks, their ability to perceive and understand human faces is rarely explored. In this work, we comprehensively evaluate existing MLLMs on face perception tasks. The quantitative results reveal that existing MLLMs struggle to handle these tasks. The primary reason is the lack of image-text datasets that contain fine-grained descriptions of human faces. To tackle this problem, we design a practical pipeline for constructing datasets, upon which we further build a novel multimodal large face perception model, namely Face-MLLM. Specifically, we re-annotate LAION-Face dataset with more detailed face captions and facial attribute labels. Besides, we re-formulate traditional face datasets using the question-answer style, which is fit for MLLMs. Together with these enriched datasets, we develop a novel three-stage MLLM training method. In the first two stages, our model learns visual-text alignment and basic visual question answering capability, respectively. In the third stage, our model learns to handle multiple specialized face perception tasks. Experimental results show that our model surpasses previous MLLMs on five famous face perception tasks. Besides, on our newly introduced zero-shot facial attribute analysis task, our Face-MLLM also presents superior performance.

Recently, large-scale diffusion models have made impressive progress in text-to-image (T2I) generation. To further equip these T2I models with fine-grained spatial control, approaches like ControlNet introduce an extra network that learns to follow a condition image. However, for every single condition type, ControlNet requires independent training on millions of data pairs with hundreds of GPU hours, which is quite expensive and makes it challenging for ordinary users to explore and develop new types of conditions. To address this problem, we propose the CtrLoRA framework, which trains a Base ControlNet to learn the common knowledge of image-to-image generation from multiple base conditions, along with condition-specific LoRAs to capture distinct characteristics of each condition. Utilizing our pretrained Base ControlNet, users can easily adapt it to new conditions, requiring as few as 1,000 data pairs and less than one hour of single-GPU training to obtain satisfactory results in most scenarios. Moreover, our CtrLoRA reduces the learnable parameters by 90% compared to ControlNet, significantly lowering the threshold to distribute and deploy the model weights. Extensive experiments on various types of conditions demonstrate the efficiency and effectiveness of our method. Codes and model weights will be released at https://github.com/xyfJASON/ctrlora.

The significant advancements in visual understanding and instruction following from Multimodal Large Language Models (MLLMs) have opened up more possibilities for broader applications in diverse and universal human-centric scenarios. However, existing image-text data may not support the precise modality alignment and integration of multi-grained information, which is crucial for human-centric visual understanding. In this paper, we introduce HERM-Bench, a benchmark for evaluating the human-centric understanding capabilities of MLLMs. Our work reveals the limitations of existing MLLMs in understanding complex human-centric scenarios. To address these challenges, we present HERM-100K, a comprehensive dataset with multi-level human-centric annotations, aimed at enhancing MLLMs' training. Furthermore, we develop HERM-7B, a MLLM that leverages enhanced training data from HERM-100K. Evaluations on HERM-Bench demonstrate that HERM-7B significantly outperforms existing MLLMs across various human-centric dimensions, reflecting the current inadequacy of data annotations used in MLLM training for human-centric visual understanding. This research emphasizes the importance of specialized datasets and benchmarks in advancing the MLLMs' capabilities for human-centric understanding.

Face Forgery Detection (FFD), or Deepfake detection, aims to determine whether a digital face is real or fake. Due to different face synthesis algorithms with diverse forgery patterns, FFD models often overfit specific patterns in training datasets, resulting in poor generalization to other unseen forgeries. This severe challenge requires FFD models to possess strong capabilities in representing complex facial features and extracting subtle forgery cues. Although previous FFD models directly employ existing backbones to represent and extract facial forgery cues, the critical role of backbones is often overlooked, particularly as their knowledge and capabilities are insufficient to address FFD challenges, inevitably limiting generalization. Therefore, it is essential to integrate the backbone pre-training configurations and seek practical solutions by revisiting the complete FFD workflow, from backbone pre-training and fine-tuning to inference of discriminant results. Specifically, we analyze the crucial contributions of backbones with different configurations in FFD task and propose leveraging the ViT network with self-supervised learning on real-face datasets to pre-train a backbone, equipping it with superior facial representation capabilities. We then build a competitive backbone fine-tuning framework that strengthens the backbone's ability to extract diverse forgery cues within a competitive learning mechanism. Moreover, we devise a threshold optimization mechanism that utilizes prediction confidence to improve the inference reliability. Comprehensive experiments demonstrate that our FFD model with the elaborate backbone achieves excellent performance in FFD and extra face-related tasks, i.e., presentation attack detection. Code and models are available at https://github.com/zhenglab/FFDBackbone.

Dynamic facial expression recognition (DFER) infers emotions from the temporal evolution of expressions, unlike static facial expression recognition (SFER), which relies solely on a single snapshot. This temporal analysis provides richer information and promises greater recognition capability. However, current DFER methods often exhibit unsatisfied performance largely due to fewer training samples compared to SFER. Given the inherent correlation between static and dynamic expressions, we hypothesize that leveraging the abundant SFER data can enhance DFER. To this end, we propose Static-for-Dynamic (S4D), a unified dual-modal learning framework that integrates SFER data as a complementary resource for DFER. Specifically, S4D employs dual-modal self-supervised pre-training on facial images and videos using a shared Vision Transformer (ViT) encoder-decoder architecture, yielding improved spatiotemporal representations. The pre-trained encoder is then fine-tuned on static and dynamic expression datasets in a multi-task learning setup to facilitate emotional information interaction. Unfortunately, vanilla multi-task learning in our study results in negative transfer. To address this, we propose an innovative Mixture of Adapter Experts (MoAE) module that facilitates task-specific knowledge acquisition while effectively extracting shared knowledge from both static and dynamic expression data. Extensive experiments demonstrate that S4D achieves a deeper understanding of DFER, setting new state-of-the-art performance on FERV39K, MAFW, and DFEW benchmarks, with weighted average recall (WAR) of 53.65\%, 58.44\%, and 76.68\%, respectively. Additionally, a systematic correlation analysis between SFER and DFER tasks is presented, which further elucidates the potential benefits of leveraging SFER., Comment: The code and model are publicly available here https://github.com/MSA-LMC/S4D

The recent wave of foundation models has witnessed tremendous success in computer vision (CV) and beyond, with the segment anything model (SAM) having sparked a passion for exploring task-agnostic visual foundation models. Empowered by its remarkable zero-shot generalization, SAM is currently challenging numerous traditional paradigms in CV, delivering extraordinary performance not only in various image segmentation and multi-modal segmentation (\eg, text-to-mask) tasks, but also in the video domain. Additionally, the latest released SAM 2 is once again sparking research enthusiasm in the realm of promptable visual segmentation for both images and videos. However, existing surveys mainly focus on SAM in various image processing tasks, a comprehensive and in-depth review in the video domain is notably absent. To address this gap, this work conducts a systematic review on SAM for videos in the era of foundation models. As the first to review the progress of SAM for videos, this work focuses on its applications to various tasks by discussing its recent advances, and innovation opportunities of developing foundation models on broad applications. We begin with a brief introduction to the background of SAM and video-related research domains. Subsequently, we present a systematic taxonomy that categorizes existing methods into three key areas: video understanding, video generation, and video editing, analyzing and summarizing their advantages and limitations. Furthermore, comparative results of SAM-based and current state-of-the-art methods on representative benchmarks, as well as insightful analysis are offered. Finally, we discuss the challenges faced by current research and envision several future research directions in the field of SAM for video and beyond., Comment: https://github.com/983632847/SAM-for-Videos

While text-to-image diffusion models demonstrate impressive generation capabilities, they also exhibit vulnerability to backdoor attacks, which involve the manipulation of model outputs through malicious triggers. In this paper, for the first time, we propose a comprehensive defense method named T2IShield to detect, localize, and mitigate such attacks. Specifically, we find the "Assimilation Phenomenon" on the cross-attention maps caused by the backdoor trigger. Based on this key insight, we propose two effective backdoor detection methods: Frobenius Norm Threshold Truncation and Covariance Discriminant Analysis. Besides, we introduce a binary-search approach to localize the trigger within a backdoor sample and assess the efficacy of existing concept editing methods in mitigating backdoor attacks. Empirical evaluations on two advanced backdoor attack scenarios show the effectiveness of our proposed defense method. For backdoor sample detection, T2IShield achieves a detection F1 score of 88.9$\%$ with low computational cost. Furthermore, T2IShield achieves a localization F1 score of 86.4$\%$ and invalidates 99$\%$ poisoned samples. Codes are released at https://github.com/Robin-WZQ/T2IShield., Comment: Accepted by ECCV2024

Currently many benchmarks have been proposed to evaluate the perception ability of the Large Vision-Language Models (LVLMs). However, most benchmarks conduct questions by selecting images from existing datasets, resulting in the potential data leakage. Besides, these benchmarks merely focus on evaluating LVLMs on the realistic style images and clean scenarios, leaving the multi-stylized images and noisy scenarios unexplored. In response to these challenges, we propose a dynamic and scalable benchmark named Dysca for evaluating LVLMs by leveraging synthesis images. Specifically, we leverage Stable Diffusion and design a rule-based method to dynamically generate novel images, questions and the corresponding answers. We consider 51 kinds of image styles and evaluate the perception capability in 20 subtasks. Moreover, we conduct evaluations under 4 scenarios (i.e., Clean, Corruption, Print Attacking and Adversarial Attacking) and 3 question types (i.e., Multi-choices, True-or-false and Free-form). Thanks to the generative paradigm, Dysca serves as a scalable benchmark for easily adding new subtasks and scenarios. A total of 24 advanced open-source LVLMs and 2 close-source LVLMs are evaluated on Dysca, revealing the drawbacks of current LVLMs. The benchmark is released at \url{https://github.com/Robin-WZQ/Dysca}., Comment: Accepted by ICLR2025

Despite the rapid progress and outstanding performance of Large Vision-Language Models (LVLMs) in recent years, LVLMs have been plagued by the issue of hallucination, i.e., LVLMs tend to generate responses that are inconsistent with the corresponding visual inputs. To evaluate the degree of hallucination in LVLMs, previous works have proposed a series of benchmarks featuring different types of tasks and evaluation metrics. However, we find that the quality of the existing hallucination benchmarks varies, with some suffering from problems, e.g., inconsistent evaluation results under repeated tests, and misalignment with human evaluation. To this end, we propose a Hallucination benchmark Quality Measurement framework (HQM), which leverages various indicators to assess the reliability and validity of existing hallucination benchmarks separately. Specifically, for reliability we explore test-retest reliability and parallel-forms reliability, while for validity we examine criterion validity and coverage of hallucination types. Furthermore, based on the results of our quality measurement, we construct a High-Quality Hallucination Benchmark (HQH) for LVLMs, which demonstrates superior reliability and validity under our HQM framework. We conduct an extensive evaluation of over 10 representative LVLMs, including GPT-4o and Gemini-1.5-Pro, to provide an in-depth analysis of the hallucination issues in existing models. Our benchmark is publicly available at https://github.com/HQHBench/HQHBench.

The emergence of Large Vision-Language Models (LVLMs) marks significant strides towards achieving general artificial intelligence. However, these advancements are accompanied by concerns about biased outputs, a challenge that has yet to be thoroughly explored. Existing benchmarks are not sufficiently comprehensive in evaluating biases due to their limited data scale, single questioning format and narrow sources of bias. To address this problem, we introduce VLBiasBench, a comprehensive benchmark designed to evaluate biases in LVLMs. VLBiasBench, features a dataset that covers nine distinct categories of social biases, including age, disability status, gender, nationality, physical appearance, race, religion, profession, social economic status, as well as two intersectional bias categories: race x gender and race x social economic status. To build a large-scale dataset, we use Stable Diffusion XL model to generate 46,848 high-quality images, which are combined with various questions to creat 128,342 samples. These questions are divided into open-ended and close-ended types, ensuring thorough consideration of bias sources and a comprehensive evaluation of LVLM biases from multiple perspectives. We conduct extensive evaluations on 15 open-source models as well as two advanced closed-source models, yielding new insights into the biases present in these models. Our benchmark is available at https://github.com/Xiangkui-Cao/VLBiasBench.

Most existing out-of-distribution (OOD) detection benchmarks classify samples with novel labels as the OOD data. However, some marginal OOD samples actually have close semantic contents to the in-distribution (ID) sample, which makes determining the OOD sample a Sorites Paradox. In this paper, we construct a benchmark named Incremental Shift OOD (IS-OOD) to address the issue, in which we divide the test samples into subsets with different semantic and covariate shift degrees relative to the ID dataset. The data division is achieved through a shift measuring method based on our proposed Language Aligned Image feature Decomposition (LAID). Moreover, we construct a Synthetic Incremental Shift (Syn-IS) dataset that contains high-quality generated images with more diverse covariate contents to complement the IS-OOD benchmark. We evaluate current OOD detection methods on our benchmark and find several important insights: (1) The performance of most OOD detection methods significantly improves as the semantic shift increases; (2) Some methods like GradNorm may have different OOD detection mechanisms as they rely less on semantic shifts to make decisions; (3) Excessive covariate shifts in the image are also likely to be considered as OOD for some methods. Our code and data are released in https://github.com/qqwsad5/IS-OOD., Comment: Accepted at NeurIPS 2024

Traditional semi-supervised learning (SSL) assumes that the feature distributions of labeled and unlabeled data are consistent which rarely holds in realistic scenarios. In this paper, we propose a novel SSL setting, where unlabeled samples are drawn from a mixed distribution that deviates from the feature distribution of labeled samples. Under this setting, previous SSL methods tend to predict wrong pseudo-labels with the model fitted on labeled data, resulting in noise accumulation. To tackle this issue, we propose Self-Supervised Feature Adaptation (SSFA), a generic framework for improving SSL performance when labeled and unlabeled data come from different distributions. SSFA decouples the prediction of pseudo-labels from the current model to improve the quality of pseudo-labels. Particularly, SSFA incorporates a self-supervised task into the SSL framework and uses it to adapt the feature extractor of the model to the unlabeled data. In this way, the extracted features better fit the distribution of unlabeled data, thereby generating high-quality pseudo-labels. Extensive experiments show that our proposed SSFA is applicable to various pseudo-label-based SSL learners and significantly improves performance in labeled, unlabeled, and even unseen distributions., Comment: 10 pages; Accepted by NeurIPS 2023

Text-to-image diffusion models, such as Stable Diffusion, generate highly realistic images from text descriptions. However, the generation of certain content at such high quality raises concerns. A prominent issue is the accurate depiction of identifiable facial images, which could lead to malicious deepfake generation and privacy violations. In this paper, we propose Anonymization Prompt Learning (APL) to address this problem. Specifically, we train a learnable prompt prefix for text-to-image diffusion models, which forces the model to generate anonymized facial identities, even when prompted to produce images of specific individuals. Extensive quantitative and qualitative experiments demonstrate the successful anonymization performance of APL, which anonymizes any specific individuals without compromising the quality of non-identity-specific image generation. Furthermore, we reveal the plug-and-play property of the learned prompt prefix, enabling its effective application across different pretrained text-to-image models for transferrable privacy and security protection against the risks of deepfakes., Comment: 15 pages, 8 figures and 5 tables

This paper presents M$^3$GPT, an advanced $\textbf{M}$ultimodal, $\textbf{M}$ultitask framework for $\textbf{M}$otion comprehension and generation. M$^3$GPT operates on three fundamental principles. The first focuses on creating a unified representation space for various motion-relevant modalities. We employ discrete vector quantization for multimodal conditional signals, such as text, music and motion/dance, enabling seamless integration into a large language model (LLM) with a single vocabulary. The second involves modeling motion generation directly in the raw motion space. This strategy circumvents the information loss associated with a discrete tokenizer, resulting in more detailed and comprehensive motion generation. Third, M$^3$GPT learns to model the connections and synergies among various motion-relevant tasks. Text, the most familiar and well-understood modality for LLMs, is utilized as a bridge to establish connections between different motion tasks, facilitating mutual reinforcement. To our knowledge, M$^3$GPT is the first model capable of comprehending and generating motions based on multiple signals. Extensive experiments highlight M$^3$GPT's superior performance across various motion-relevant tasks and its powerful zero-shot generalization capabilities for extremely challenging tasks. Project page: \url{https://github.com/luomingshuang/M3GPT}., Comment: Accepted at NeurIPS 2024, 21 pages, 7 figures

The visual pathway of human brain includes two sub-pathways, ie, the ventral pathway and the dorsal pathway, which focus on object identification and dynamic information modeling, respectively. Both pathways comprise multi-layer structures, with each layer responsible for processing different aspects of visual information. Inspired by visual information processing mechanism of the human brain, we propose the Brain Inspired Masked Modeling (BIMM) framework, aiming to learn comprehensive representations from videos. Specifically, our approach consists of ventral and dorsal branches, which learn image and video representations, respectively. Both branches employ the Vision Transformer (ViT) as their backbone and are trained using masked modeling method. To achieve the goals of different visual cortices in the brain, we segment the encoder of each branch into three intermediate blocks and reconstruct progressive prediction targets with light weight decoders. Furthermore, drawing inspiration from the information-sharing mechanism in the visual pathways, we propose a partial parameter sharing strategy between the branches during training. Extensive experiments demonstrate that BIMM achieves superior performance compared to the state-of-the-art methods.

Although face analysis has achieved remarkable improvements in the past few years, designing a multi-task face analysis model is still challenging. Most face analysis tasks are studied as separate problems and do not benefit from the synergy among related tasks. In this work, we propose a novel task-adaptive multi-task face analysis method named as Q-Face, which simultaneously performs multiple face analysis tasks with a unified model. We fuse the features from multiple layers of a large-scale pre-trained model so that the whole model can use both local and global facial information to support multiple tasks. Furthermore, we design a task-adaptive module that performs cross-attention between a set of query vectors and the fused multi-stage features and finally adaptively extracts desired features for each face analysis task. Extensive experiments show that our method can perform multiple tasks simultaneously and achieves state-of-the-art performance on face expression recognition, action unit detection, face attribute analysis, age estimation, and face pose estimation. Compared to conventional methods, our method opens up new possibilities for multi-task face analysis and shows the potential for both accuracy and efficiency., Comment: Ever submitted to ECCV2024

Few-shot segmentation (FSS) aims to train a model which can segment the object from novel classes with a few labeled samples. The insufficient generalization ability of models leads to unsatisfactory performance when the models lack enough labeled data from the novel classes. Considering that there are abundant unlabeled data available, it is promising to improve the generalization ability by exploiting these various data. For leveraging unlabeled data, we propose a novel method, named Image to Pseudo-Episode (IPE), to generate pseudo-episodes from unlabeled data. Specifically, our method contains two modules, i.e., the pseudo-label generation module and the episode generation module. The former module generates pseudo-labels from unlabeled images by the spectral clustering algorithm, and the latter module generates pseudo-episodes from pseudo-labeled images by data augmentation methods. Extensive experiments on PASCAL-$5^i$ and COCO-$20^i$ demonstrate that our method achieves the state-of-the-art performance for FSS.

Current clothes-changing person re-identification (re-id) approaches usually perform retrieval based on clothes-irrelevant features, while neglecting the potential of clothes-relevant features. However, we observe that relying solely on clothes-irrelevant features for clothes-changing re-id is limited, since they often lack adequate identity information and suffer from large intra-class variations. On the contrary, clothes-relevant features can be used to discover same-clothes intermediaries that possess informative identity clues. Based on this observation, we propose a Feasibility-Aware Intermediary Matching (FAIM) framework to additionally utilize clothes-relevant features for retrieval. Firstly, an Intermediary Matching (IM) module is designed to perform an intermediary-assisted matching process. This process involves using clothes-relevant features to find informative intermediates, and then using clothes-irrelevant features of these intermediates to complete the matching. Secondly, in order to reduce the negative effect of low-quality intermediaries, an Intermediary-Based Feasibility Weighting (IBFW) module is designed to evaluate the feasibility of intermediary matching process by assessing the quality of intermediaries. Extensive experiments demonstrate that our method outperforms state-of-the-art methods on several widely-used clothes-changing re-id benchmarks.

Predicting the trajectories of road agents is essential for autonomous driving systems. The recent mainstream methods follow a static paradigm, which predicts the future trajectory by using a fixed duration of historical frames. These methods make the predictions independently even at adjacent time steps, which leads to potential instability and temporal inconsistency. As successive time steps have largely overlapping historical frames, their forecasting should have intrinsic correlation, such as overlapping predicted trajectories should be consistent, or be different but share the same motion goal depending on the road situation. Motivated by this, in this work, we introduce HPNet, a novel dynamic trajectory forecasting method. Aiming for stable and accurate trajectory forecasting, our method leverages not only historical frames including maps and agent states, but also historical predictions. Specifically, we newly design a Historical Prediction Attention module to automatically encode the dynamic relationship between successive predictions. Besides, it also extends the attention range beyond the currently visible window benefitting from the use of historical predictions. The proposed Historical Prediction Attention together with the Agent Attention and Mode Attention is further formulated as the Triple Factorized Attention module, serving as the core design of HPNet.Experiments on the Argoverse and INTERACTION datasets show that HPNet achieves state-of-the-art performance, and generates accurate and stable future trajectories. Our code are available at https://github.com/XiaolongTang23/HPNet., Comment: CVPR2024

As XR technology continues to advance rapidly, 3D generation and editing are increasingly crucial. Among these, stylization plays a key role in enhancing the appearance of 3D models. By utilizing stylization, users can achieve consistent artistic effects in 3D editing using a single reference style image, making it a user-friendly editing method. However, recent NeRF-based 3D stylization methods encounter efficiency issues that impact the user experience, and their implicit nature limits their ability to accurately transfer geometric pattern styles. Additionally, the ability for artists to apply flexible control over stylized scenes is considered highly desirable to foster an environment conducive to creative exploration. To address the above issues, we introduce StylizedGS, an efficient 3D neural style transfer framework with adaptable control over perceptual factors based on 3D Gaussian Splatting (3DGS) representation. We propose a filter-based refinement to eliminate floaters that affect the stylization effects in the scene reconstruction process. The nearest neighbor-based style loss is introduced to achieve stylization by fine-tuning the geometry and color parameters of 3DGS, while a depth preservation loss with other regularizations is proposed to prevent the tampering of geometry content. Moreover, facilitated by specially designed losses, StylizedGS enables users to control color, stylized scale, and regions during the stylization to possess customization capabilities. Our method achieves high-quality stylization results characterized by faithful brushstrokes and geometric consistency with flexible controls. Extensive experiments across various scenes and styles demonstrate the effectiveness and efficiency of our method concerning both stylization quality and inference speed.

Multimodal large language models (MLLMs) are designed to process and integrate information from multiple sources, such as text, speech, images, and videos. Despite its success in language understanding, it is critical to evaluate the performance of downstream tasks for better human-centric applications. This paper assesses the application of MLLMs with 5 crucial abilities for affective computing, spanning from visual affective tasks and reasoning tasks. The results show that \gpt has high accuracy in facial action unit recognition and micro-expression detection while its general facial expression recognition performance is not accurate. We also highlight the challenges of achieving fine-grained micro-expression recognition and the potential for further study and demonstrate the versatility and potential of \gpt for handling advanced tasks in emotion recognition and related fields by integrating with task-related agents for more complex tasks, such as heart rate estimation through signal processing. In conclusion, this paper provides valuable insights into the potential applications and challenges of MLLMs in human-centric computing. Our interesting examples are at https://github.com/EnVision-Research/GPT4Affectivity.

Facial action unit (AU) detection, aiming to classify AU present in the facial image, has long suffered from insufficient AU annotations. In this paper, we aim to mitigate this data scarcity issue by learning AU representations from a large number of unlabelled facial videos in a contrastive learning paradigm. We formulate the self-supervised AU representation learning signals in two-fold: (1) AU representation should be frame-wisely discriminative within a short video clip; (2) Facial frames sampled from different identities but show analogous facial AUs should have consistent AU representations. As to achieve these goals, we propose to contrastively learn the AU representation within a video clip and devise a cross-identity reconstruction mechanism to learn the person-independent representations. Specially, we adopt a margin-based temporal contrastive learning paradigm to perceive the temporal AU coherence and evolution characteristics within a clip that consists of consecutive input facial frames. Moreover, the cross-identity reconstruction mechanism facilitates pushing the faces from different identities but show analogous AUs close in the latent embedding space. Experimental results on three public AU datasets demonstrate that the learned AU representation is discriminative for AU detection. Our method outperforms other contrastive learning methods and significantly closes the performance gap between the self-supervised and supervised AU detection approaches.

Previous Face Anti-spoofing (FAS) methods face the challenge of generalizing to unseen domains, mainly because most existing FAS datasets are relatively small and lack data diversity. Thanks to the development of face recognition in the past decade, numerous real face images are available publicly, which are however neglected previously by the existing literature. In this paper, we propose an Anomalous cue Guided FAS (AG-FAS) method, which can effectively leverage large-scale additional real faces for improving model generalization via a De-fake Face Generator (DFG). Specifically, by training on a large-scale real face only dataset, the generator obtains the knowledge of what a real face should be like, and thus has the capability of generating a "real" version of any input face image. Consequently, the difference between the input face and the generated "real" face can be treated as cues of attention for the fake feature learning. With the above ideas, an Off-real Attention Network (OA-Net) is proposed which allocates its attention to the spoof region of the input according to the anomalous cue. Extensive experiments on a total of nine public datasets show our method achieves state-of-the-art results under cross-domain evaluations with unseen scenarios and unknown presentation attacks. Besides, we provide theoretical analysis demonstrating the effectiveness of the proposed anomalous cues., Comment: 14 pages

Large-scale pre-trained vision-language models like CLIP have demonstrated impressive performance across various tasks, and exhibit remarkable zero-shot generalization capability, while they are also vulnerable to imperceptible adversarial examples. Existing works typically employ adversarial training (fine-tuning) as a defense method against adversarial examples. However, direct application to the CLIP model may result in overfitting, compromising the model's capacity for generalization. In this paper, we propose Pre-trained Model Guided Adversarial Fine-Tuning (PMG-AFT) method, which leverages supervision from the original pre-trained model by carefully designing an auxiliary branch, to enhance the model's zero-shot adversarial robustness. Specifically, PMG-AFT minimizes the distance between the features of adversarial examples in the target model and those in the pre-trained model, aiming to preserve the generalization features already captured by the pre-trained model. Extensive Experiments on 15 zero-shot datasets demonstrate that PMG-AFT significantly outperforms the state-of-the-art method, improving the top-1 robust accuracy by an average of 4.99%. Furthermore, our approach consistently improves clean accuracy by an average of 8.72%. Our code is available at https://github.com/serendipity1122/Pre-trained-Model-Guided-Fine-Tuning-for-Zero-Shot-Adversarial-Robustness., Comment: Accepted by CVPR 2024

In recent years, the Vision Transformer (ViT) model has gradually become mainstream in various computer vision tasks, and the robustness of the model has received increasing attention. However, existing large models tend to prioritize performance during training, potentially neglecting the robustness, which may lead to serious security concerns. In this paper, we establish a new challenge: exploring how to use a small number of additional parameters for adversarial finetuning to quickly and effectively enhance the adversarial robustness of a standardly trained model. To address this challenge, we develop the novel LNLoRA module, incorporating a learnable layer normalization before the conventional LoRA module, which helps mitigate magnitude differences in parameters between the adversarial and standard training paradigms. Furthermore, we propose the FullLoRA-AT framework by integrating the learnable LNLoRA modules into all key components of ViT-based models while keeping the pretrained model frozen, which can significantly improve the model robustness via adversarial finetuning in a parameter-efficient manner. Extensive experiments on CIFAR-10, CIFAR-100, and Imagenette demonstrate the superiority of our proposed FullLoRA-AT framework. It achieves comparable robustness with full finetuning while only requiring about 5% of the learnable parameters. This also effectively addresses concerns regarding extra model storage space and enormous training time caused by adversarial finetuning., Comment: 10 pages, 2 figures, 6 tables

We present a unified, promptable model capable of simultaneously segmenting, recognizing, and captioning anything. Unlike SAM, we aim to build a versatile region representation in the wild via visual prompting. To achieve this, we train a generalizable model with massive segmentation masks, \eg, SA-1B masks, and semantic priors from a pre-trained CLIP model with 5 billion parameters. Specifically, we construct a promptable image decoder by adding a semantic token to each mask token. The semantic token is responsible for learning the semantic priors in a predefined concept space. Through joint optimization of segmentation on mask tokens and concept prediction on semantic tokens, our model exhibits strong regional recognition and localization capabilities. For example, an additional 38M-parameter causal text decoder trained from scratch sets a new record with a CIDEr score of 164.7 on the Visual Genome region captioning task. We believe this model can be a versatile region-level image tokenizer, capable of encoding general-purpose region context for a broad range of visual perception tasks. Code and models are available at {\footnotesize \url{https://github.com/baaivision/tokenize-anything}}., Comment: code, model, and demo: https://github.com/baaivision/tokenize-anything

Dynamic facial expression recognition (DFER) in the wild is still hindered by data limitations, e.g., insufficient quantity and diversity of pose, occlusion and illumination, as well as the inherent ambiguity of facial expressions. In contrast, static facial expression recognition (SFER) currently shows much higher performance and can benefit from more abundant high-quality training data. Moreover, the appearance features and dynamic dependencies of DFER remain largely unexplored. To tackle these challenges, we introduce a novel Static-to-Dynamic model (S2D) that leverages existing SFER knowledge and dynamic information implicitly encoded in extracted facial landmark-aware features, thereby significantly improving DFER performance. Firstly, we build and train an image model for SFER, which incorporates a standard Vision Transformer (ViT) and Multi-View Complementary Prompters (MCPs) only. Then, we obtain our video model (i.e., S2D), for DFER, by inserting Temporal-Modeling Adapters (TMAs) into the image model. MCPs enhance facial expression features with landmark-aware features inferred by an off-the-shelf facial landmark detector. And the TMAs capture and model the relationships of dynamic changes in facial expressions, effectively extending the pre-trained image model for videos. Notably, MCPs and TMAs only increase a fraction of trainable parameters (less than +10\%) to the original image model. Moreover, we present a novel Emotion-Anchors (i.e., reference samples for each emotion category) based Self-Distillation Loss to reduce the detrimental influence of ambiguous emotion labels, further enhancing our S2D. Experiments conducted on popular SFER and DFER datasets show that we achieve the state of the art., Comment: The code is available at https://github.com/MSA-LMC/S2D. The paper has been accepted for publication in IEEE Transactions on Affective Computing (TAFFC)

In this work, we focus on leveraging facial cues beyond the lip region for robust Audio-Visual Speech Enhancement (AVSE). The facial region, encompassing the lip region, reflects additional speech-related attributes such as gender, skin color, nationality, etc., which contribute to the effectiveness of AVSE. However, static and dynamic speech-unrelated attributes also exist, causing appearance changes during speech. To address these challenges, we propose a Dual Attention Cooperative Framework, DualAVSE, to ignore speech-unrelated information, capture speech-related information with facial cues, and dynamically integrate it with the audio signal for AVSE. Specifically, we introduce a spatial attention-based visual encoder to capture and enhance visual speech information beyond the lip region, incorporating global facial context and automatically ignoring speech-unrelated information for robust visual feature extraction. Additionally, a dynamic visual feature fusion strategy is introduced by integrating a temporal-dimensional self-attention module, enabling the model to robustly handle facial variations. The acoustic noise in the speaking process is variable, impacting audio quality. Therefore, a dynamic fusion strategy for both audio and visual features is introduced to address this issue. By integrating cooperative dual attention in the visual encoder and audio-visual fusion strategy, our model effectively extracts beneficial speech information from both audio and visual cues for AVSE. Thorough analysis and comparison on different datasets, including normal and challenging cases with unreliable or absent visual information, consistently show our model outperforming existing methods across multiple metrics., Comment: Accepted to BMVC 2023 15 pages, 2 figures

In this paper, we propose a novel method for speaker adaptation in lip reading, motivated by two observations. Firstly, a speaker's own characteristics can always be portrayed well by his/her few facial images or even a single image with shallow networks, while the fine-grained dynamic features associated with speech content expressed by the talking face always need deep sequential networks to represent accurately. Therefore, we treat the shallow and deep layers differently for speaker adaptive lip reading. Secondly, we observe that a speaker's unique characteristics ( e.g. prominent oral cavity and mandible) have varied effects on lip reading performance for different words and pronunciations, necessitating adaptive enhancement or suppression of the features for robust lip reading. Based on these two observations, we propose to take advantage of the speaker's own characteristics to automatically learn separable hidden unit contributions with different targets for shallow layers and deep layers respectively. For shallow layers where features related to the speaker's characteristics are stronger than the speech content related features, we introduce speaker-adaptive features to learn for enhancing the speech content features. For deep layers where both the speaker's features and the speech content features are all expressed well, we introduce the speaker-adaptive features to learn for suppressing the speech content irrelevant noise for robust lip reading. Our approach consistently outperforms existing methods, as confirmed by comprehensive analysis and comparison across different settings. Besides the evaluation on the popular LRW-ID and GRID datasets, we also release a new dataset for evaluation, CAS-VSR-S68h, to further assess the performance in an extreme setting where just a few speakers are available but the speech content covers a large and diversified range., Comment: Accepted to BMVC 2023 20pages

Learning generalizable representation and classifier for class-imbalanced data is challenging for data-driven deep models. Most studies attempt to re-balance the data distribution, which is prone to overfitting on tail classes and underfitting on head classes. In this work, we propose Dual Compensation Residual Networks to better fit both tail and head classes. Firstly, we propose dual Feature Compensation Module (FCM) and Logit Compensation Module (LCM) to alleviate the overfitting issue. The design of these two modules is based on the observation: an important factor causing overfitting is that there is severe feature drift between training and test data on tail classes. In details, the test features of a tail category tend to drift towards feature cloud of multiple similar head categories. So FCM estimates a multi-mode feature drift direction for each tail category and compensate for it. Furthermore, LCM translates the deterministic feature drift vector estimated by FCM along intra-class variations, so as to cover a larger effective compensation space, thereby better fitting the test features. Secondly, we propose a Residual Balanced Multi-Proxies Classifier (RBMC) to alleviate the under-fitting issue. Motivated by the observation that re-balancing strategy hinders the classifier from learning sufficient head knowledge and eventually causes underfitting, RBMC utilizes uniform learning with a residual path to facilitate classifier learning. Comprehensive experiments on Long-tailed and Class-Incremental benchmarks validate the efficacy of our method., Comment: 20 pages

Vision Transformers have achieved great success in computer visions, delivering exceptional performance across various tasks. However, their inherent reliance on sequential input enforces the manual partitioning of images into patch sequences, which disrupts the image's inherent structural and semantic continuity. To handle this, we propose a novel Pattern Transformer (Patternformer) to adaptively convert images to pattern sequences for Transformer input. Specifically, we employ the Convolutional Neural Network to extract various patterns from the input image, with each channel representing a unique pattern that is fed into the succeeding Transformer as a visual token. By enabling the network to optimize these patterns, each pattern concentrates on its local region of interest, thereby preserving its intrinsic structural and semantic information. Only employing the vanilla ResNet and Transformer, we have accomplished state-of-the-art performance on CIFAR-10 and CIFAR-100, and have achieved competitive results on ImageNet.

In Knowledge Distillation, the teacher is generally much larger than the student, making the solution of the teacher likely to be difficult for the student to learn. To ease the mimicking difficulty, we introduce a triplet knowledge distillation mechanism named TriKD. Besides teacher and student, TriKD employs a third role called anchor model. Before distillation begins, the pre-trained anchor model delimits a subspace within the full solution space of the target problem. Solutions within the subspace are expected to be easy targets that the student could mimic well. Distillation then begins in an online manner, and the teacher is only allowed to express solutions within the aforementioned subspace. Surprisingly, benefiting from accurate but easy-to-mimic hints, the student can finally perform well. After the student is well trained, it can be used as the new anchor for new students, forming a curriculum learning strategy. Our experiments on image classification and face recognition with various models clearly demonstrate the effectiveness of our method. Furthermore, the proposed TriKD is also effective in dealing with the overfitting issue. Moreover, our theoretical analysis supports the rationality of our triplet distillation.

Feature distillation makes the student mimic the intermediate features of the teacher. Nearly all existing feature-distillation methods use L2 distance or its slight variants as the distance metric between teacher and student features. However, while L2 distance is isotropic w.r.t. all dimensions, the neural network's operation on different dimensions is usually anisotropic, i.e., perturbations with the same 2-norm but in different dimensions of intermediate features lead to changes in the final output with largely different magnitude. Considering this, we argue that the similarity between teacher and student features should not be measured merely based on their appearance (i.e., L2 distance), but should, more importantly, be measured by their difference in function, namely how later layers of the network will read, decode, and process them. Therefore, we propose Function-Consistent Feature Distillation (FCFD), which explicitly optimizes the functional similarity between teacher and student features. The core idea of FCFD is to make teacher and student features not only numerically similar, but more importantly produce similar outputs when fed to the later part of the same network. With FCFD, the student mimics the teacher more faithfully and learns more from the teacher. Extensive experiments on image classification and object detection demonstrate the superiority of FCFD to existing methods. Furthermore, we can combine FCFD with many existing methods to obtain even higher accuracy. Our codes are available at https://github.com/LiuDongyang6/FCFD., Comment: ICLR 2023

With the development of deep generative models, recent years have seen great success of Chinese landscape painting generation. However, few works focus on controllable Chinese landscape painting generation due to the lack of data and limited modeling capabilities. In this work, we propose a controllable Chinese landscape painting generation method named CCLAP, which can generate painting with specific content and style based on Latent Diffusion Model. Specifically, it consists of two cascaded modules, i.e., content generator and style aggregator. The content generator module guarantees the content of generated paintings specific to the input text. While the style aggregator module is to generate paintings of a style corresponding to a reference image. Moreover, a new dataset of Chinese landscape paintings named CLAP is collected for comprehensive evaluation. Both the qualitative and quantitative results demonstrate that our method achieves state-of-the-art performance, especially in artfully-composed and artistic conception. Codes are available at https://github.com/Robin-WZQ/CCLAP., Comment: 8 pages,13 figures