Jing Shao

Zhelun Shi, Zhipin Wang, Hongxing Fan et al.

Multimodal Large Language Models (MLLMs) have shown impressive abilities in interacting with visual content with myriad potential downstream tasks. However, even though a list of benchmarks has been proposed, the capabilities and limitations of MLLMs are still not comprehensively understood, due to a lack of a standardized and holistic evaluation framework. To this end, we present the first Comprehensive Evaluation Framework (ChEF) that can holistically profile each MLLM and fairly compare different MLLMs. First, we structure ChEF as four modular components, i.e., Scenario as scalable multimodal datasets, Instruction as flexible instruction retrieving formulae, Inferencer as reliable question answering strategies, and Metric as indicative task-specific score functions. Based on them, ChEF facilitates versatile evaluations in a standardized framework, and new evaluations can be built by designing new Recipes (systematic selection of these four components). Notably, current MLLM benchmarks can be readily summarized as recipes of ChEF. Second, we introduce 6 new recipes to quantify competent MLLMs' desired capabilities (or called desiderata, i.e., calibration, in-context learning, instruction following, language performance, hallucination, and robustness) as reliable agents that can perform real-world multimodal interactions. Third, we conduct a large-scale evaluation of 9 prominent MLLMs on 9 scenarios and 6 desiderata. Our evaluation summarized over 20 valuable observations concerning the generalizability of MLLMs across various scenarios and the composite capability of MLLMs required for multimodal interactions. We will publicly release all the detailed implementations for further analysis, as well as an easy-to-use modular toolkit for the integration of new recipes and models, so that ChEF can be a growing evaluation framework for the MLLM community.

Ling Tang, Jilin Mei, Qian Chen et al.

Large language models (LLMs) can simulate human-like reasoning and decision-making in individual agents. LLM-powered multi-agent systems (MAS) combine such agents to simulate population-scale social phenomena such as polarization, information cascades, and market panics. Such studies require attributing macro emergence to individual agents, but existing axiomatic methods scale combinatorially in $N$ and have been confined to $N \lesssim 10^3$, while the phenomena they explain occur at $N \geq 10^6$. We address this gap by adapting Aumann--Shapley path-integral attribution to LLM-powered MAS at million-agent scale; the resulting method satisfies all four axioms, runs four to five orders of magnitude faster than sampled Shapley on the same hardware. We use this method to test the scale gap empirically: across 14 days of public Bluesky data ($1{,}671{,}587$ active users), we compute the attribution at both full scale and the visibility-biased $N = 10^2$ convenience sample used by small-scale studies, and the two disagree structurally. At full scale the long tail and middle tier jointly carry the majority; the biased small panel attributes almost everything to a few high-follower accounts. We then prove that under any nonlinear macro indicator the disagreement cannot be reduced by post-hoc rescaling: an Attribution Scaling Bias theorem shows that no global rescaling factor can reconcile small-scale and full-scale attribution. Full-scale attribution is therefore not a methodological choice but a theoretical requirement for any nonlinear macro indicator.

Lijun Li, Zhelun Shi, Xuhao Hu et al.

Text-to-image (T2I) models have rapidly advanced, enabling the generation of high-quality images from text prompts across various domains. However, these models present notable safety concerns, including the risk of generating harmful, biased, or private content. Current research on assessing T2I safety remains in its early stages. While some efforts have been made to evaluate models on specific safety dimensions, many critical risks remain unexplored. To address this gap, we introduce T2ISafety, a safety benchmark that evaluates T2I models across three key domains: toxicity, fairness, and bias. We build a detailed hierarchy of 12 tasks and 44 categories based on these three domains, and meticulously collect 70K corresponding prompts. Based on this taxonomy and prompt set, we build a large-scale T2I dataset with 68K manually annotated images and train an evaluator capable of detecting critical risks that previous work has failed to identify, including risks that even ultra-large proprietary models like GPTs cannot correctly detect. We evaluate 12 prominent diffusion models on T2ISafety and reveal several concerns including persistent issues with racial fairness, a tendency to generate toxic content, and significant variation in privacy protection across the models, even with defense methods like concept erasing. Data and evaluator are released under https://github.com/adwardlee/t2i_safety.

Chen Qian, Jie Zhang, Wei Yao et al.

Ensuring the trustworthiness of large language models (LLMs) is crucial. Most studies concentrate on fully pre-trained LLMs to better understand and improve LLMs' trustworthiness. In this paper, to reveal the untapped potential of pre-training, we pioneer the exploration of LLMs' trustworthiness during this period, focusing on five key dimensions: reliability, privacy, toxicity, fairness, and robustness. To begin with, we apply linear probing to LLMs. The high probing accuracy suggests that \textit{LLMs in early pre-training can already distinguish concepts in each trustworthiness dimension}. Therefore, to further uncover the hidden possibilities of pre-training, we extract steering vectors from a LLM's pre-training checkpoints to enhance the LLM's trustworthiness. Finally, inspired by~\citet{choi2023understanding} that mutual information estimation is bounded by linear probing accuracy, we also probe LLMs with mutual information to investigate the dynamics of trustworthiness during pre-training. We are the first to observe a similar two-phase phenomenon: fitting and compression~\citep{shwartz2017opening}. This research provides an initial exploration of trustworthiness modeling during LLM pre-training, seeking to unveil new insights and spur further developments in the field. We will make our code publicly accessible at \url{https://github.com/ChnQ/TracingLLM}.

Ziqi Miao, Lijun Li, Yuan Xiong et al.

Contextual priming, where earlier stimuli covertly bias later judgments, offers an unexplored attack surface for large language models (LLMs). We uncover a contextual priming vulnerability in which the previous response in the dialogue can steer its subsequent behavior toward policy-violating content. Building on this insight, we propose Response Attack, which uses an auxiliary LLM to generate a mildly harmful response to a paraphrased version of the original malicious query. They are then formatted into the dialogue and followed by a succinct trigger prompt, thereby priming the target model to generate harmful content. Across eight open-source and proprietary LLMs, RA consistently outperforms seven state-of-the-art jailbreak techniques, achieving higher attack success rates. To mitigate this threat, we construct and release a context-aware safety fine-tuning dataset, which significantly reduces the attack success rate while preserving model capabilities. The code and data are available at https://github.com/Dtc7w3PQ/Response-Attack.

Yuan Xiong, Ziqi Miao, Lijun Li et al.

While Multimodal Large Language Models (MLLMs) show remarkable capabilities, their safety alignments are susceptible to jailbreak attacks. Existing attack methods typically focus on text-image interplay, treating the visual modality as a secondary prompt. This approach underutilizes the unique potential of images to carry complex, contextual information. To address this gap, we propose a new image-centric attack method, Contextual Image Attack (CIA), which employs a multi-agent system to subtly embeds harmful queries into seemingly benign visual contexts using four distinct visualization strategies. To further enhance the attack's efficacy, the system incorporate contextual element enhancement and automatic toxicity obfuscation techniques. Experimental results on the MMSafetyBench-tiny dataset show that CIA achieves high toxicity scores of 4.73 and 4.83 against the GPT-4o and Qwen2.5-VL-72B models, respectively, with Attack Success Rates (ASR) reaching 86.31\% and 91.07\%. Our method significantly outperforms prior work, demonstrating that the visual modality itself is a potent vector for jailbreaking advanced MLLMs.

Teli Ma, Shijie Geng, Mengmeng Wang et al.

The visual world naturally exhibits a long-tailed distribution of open classes, which poses great challenges to modern visual systems. Existing approaches either perform class re-balancing strategies or directly improve network modules to address the problem. However, they still train models with a finite set of predefined labels, limiting their supervision information and restricting their transferability to novel instances. Recent advances in large-scale contrastive visual-language pretraining shed light on a new pathway for visual recognition. With open-vocabulary supervisions, pretrained contrastive vision-language models learn powerful multimodal representations that are promising to handle data deficiency and unseen concepts. By calculating the semantic similarity between visual and text inputs, visual recognition is converted to a vision-language matching problem. Inspired by this, we propose BALLAD to leverage contrastive vision-language models for long-tailed recognition. We first continue pretraining the vision-language backbone through contrastive learning on a specific long-tailed target dataset. Afterward, we freeze the backbone and further employ an additional adapter layer to enhance the representations of tail classes on balanced training samples built with re-sampling strategies. Extensive experiments have been conducted on three popular long-tailed recognition benchmarks. As a result, our simple and effective approach sets the new state-of-the-art performances and outperforms competitive baselines with a large margin. Code is released at https://github.com/gaopengcuhk/BALLAD.

Yangguang Li, Feng Liang, Lichen Zhao et al.

Recently, large-scale Contrastive Language-Image Pre-training (CLIP) has attracted unprecedented attention for its impressive zero-shot recognition ability and excellent transferability to downstream tasks. However, CLIP is quite data-hungry and requires 400M image-text pairs for pre-training, thereby restricting its adoption. This work proposes a novel training paradigm, Data efficient CLIP (DeCLIP), to alleviate this limitation. We demonstrate that by carefully utilizing the widespread supervision among the image-text pairs, our De-CLIP can learn generic visual features more efficiently. Instead of using the single image-text contrastive supervision, we fully exploit data potential through the use of (1) self-supervision within each modality; (2) multi-view supervision across modalities; (3) nearest-neighbor supervision from other similar pairs. Benefiting from intrinsic supervision, our DeCLIP-ResNet50 can achieve 60.4% zero-shot top1 accuracy on ImageNet, which is 0.8% above the CLIP-ResNet50 while using 7.1 x fewer data. Our DeCLIP-ResNet50 outperforms its counterpart in 8 out of 11 visual datasets when transferred to downstream tasks. Moreover, Scaling up the model and computing also works well in our framework.Our code, dataset and models are released at: https://github.com/Sense-GVT/DeCLIP

Siyu Chen, Junting Pan, Guanglu Song et al.

This technical report introduces our winning solution to the spatio-temporal action localization track, AVA-Kinetics Crossover, in ActivityNet Challenge 2020. Our entry is mainly based on Actor-Context-Actor Relation Network. We describe technical details for the new AVA-Kinetics dataset, together with some experimental results. Without any bells and whistles, we achieved 39.62 mAP on the test set of AVA-Kinetics, which outperforms other entries by a large margin. Code will be available at: https://github.com/Siyu-C/ACAR-Net.

Junting Pan, Siyu Chen, Mike Zheng Shou et al.

Localizing persons and recognizing their actions from videos is a challenging task towards high-level video understanding. Recent advances have been achieved by modeling direct pairwise relations between entities. In this paper, we take one step further, not only model direct relations between pairs but also take into account indirect higher-order relations established upon multiple elements. We propose to explicitly model the Actor-Context-Actor Relation, which is the relation between two actors based on their interactions with the context. To this end, we design an Actor-Context-Actor Relation Network (ACAR-Net) which builds upon a novel High-order Relation Reasoning Operator and an Actor-Context Feature Bank to enable indirect relation reasoning for spatio-temporal action localization. Experiments on AVA and UCF101-24 datasets show the advantages of modeling actor-context-actor relations, and visualization of attention maps further verifies that our model is capable of finding relevant higher-order relations to support action detection. Notably, our method ranks first in the AVA-Kineticsaction localization task of ActivityNet Challenge 2020, out-performing other entries by a significant margin (+6.71mAP). Training code and models will be available at https://github.com/Siyu-C/ACAR-Net.

Guojun Yin, Lu Sheng, Bin Liu et al.

Recognizing visual relationships <subject-predicate-object> among any pair of localized objects is pivotal for image understanding. Previous studies have shown remarkable progress in exploiting linguistic priors or external textual information to improve the performance. In this work, we investigate an orthogonal perspective based on feature interactions. We show that by encouraging deep message propagation and interactions between local object features and global predicate features, one can achieve compelling performance in recognizing complex relationships without using any linguistic priors. To this end, we present two new pooling cells to encourage feature interactions: (i) Contrastive ROI Pooling Cell, which has a unique deROI pooling that inversely pools local object features to the corresponding area of global predicate features. (ii) Pyramid ROI Pooling Cell, which broadcasts global predicate features to reinforce local object features.The two cells constitute a Spatiality-Context-Appearance Module (SCA-M), which can be further stacked consecutively to form our final Zoom-Net.We further shed light on how one could resolve ambiguous and noisy object and predicate annotations by Intra-Hierarchical trees (IH-tree). Extensive experiments conducted on Visual Genome dataset demonstrate the effectiveness of our feature-oriented approach compared to state-of-the-art methods (Acc@1 11.42% from 8.16%) that depend on explicit modeling of linguistic interactions. We further show that SCA-M can be incorporated seamlessly into existing approaches to improve the performance by a large margin. The source code will be released on https://github.com/gjyin91/ZoomNet.

Chenxin Li, Hengyu Liu, Yifan Liu et al.

Generative models hold promise for revolutionizing medical education, robot-assisted surgery, and data augmentation for machine learning. Despite progress in generating 2D medical images, the complex domain of clinical video generation has largely remained untapped.This paper introduces \model, an innovative approach to generate medical videos that simulate clinical endoscopy scenes. We present a novel generative model design that integrates a meticulously crafted spatial-temporal video transformer with advanced 2D vision foundation model priors, explicitly modeling spatial-temporal dynamics during video generation. We also pioneer the first public benchmark for endoscopy simulation with video generation models, adapting existing state-of-the-art methods for this endeavor.Endora demonstrates exceptional visual quality in generating endoscopy videos, surpassing state-of-the-art methods in extensive testing. Moreover, we explore how this endoscopy simulator can empower downstream video analysis tasks and even generate 3D medical scenes with multi-view consistency. In a nutshell, Endora marks a notable breakthrough in the deployment of generative AI for clinical endoscopy research, setting a substantial stage for further advances in medical content generation. For more details, please visit our project page: https://endora-medvidgen.github.io/.

Chenhui Zuo, Kaibo He, Jing Shao et al.

Modeling and control of the human musculoskeletal system is important for understanding human motor functions, developing embodied intelligence, and optimizing human-robot interaction systems. However, current human musculoskeletal models are restricted to a limited range of body parts and often with a reduced number of muscles. There is also a lack of algorithms capable of controlling over 600 muscles to generate reasonable human movements. To fill this gap, we build a musculoskeletal model (MS-Human-700) with 90 body segments, 206 joints, and 700 muscle-tendon units, allowing simulation of full-body dynamics and interaction with various devices. We develop a new algorithm using low-dimensional representation and hierarchical deep reinforcement learning to achieve state-of-the-art full-body control. We validate the effectiveness of our model and algorithm in simulations with real human locomotion data. The musculoskeletal model, along with its control algorithm, will be made available to the research community to promote a deeper understanding of human motion control and better design of interactive robots. Project page: https://lnsgroup.cc/research/MS-Human-700

Lehan He, Zeren Chen, Zhelun Shi et al.

The success of Direct Preference Optimization (DPO) in mitigating hallucinations in Vision Language Models (VLMs) critically hinges on the true reward gaps within preference pairs. However, current methods, typically relying on ranking or rewriting strategies, often struggle to optimize these reward gaps in a systematic way during data curation. A core difficulty lies in precisely characterizing and strategically manipulating the overall reward gap configuration, that is, the deliberate design of how to shape these reward gaps within each preference pair across the data. To address this, we introduce Topic-level Preference Rewriting(TPR), a novel framework designed for the systematic optimization of reward gap configuration. Through selectively replacing semantic topics within VLM responses with model's own resampled candidates for targeted rewriting, TPR can provide topic-level control over fine-grained semantic details. This precise control enables advanced data curation strategies, such as progressively adjusting the difficulty of rejected responses, thereby sculpting an effective reward gap configuration that guides the model to overcome challenging hallucinations. Comprehensive experiments demonstrate TPR achieves state-of-the-art performance on multiple hallucination benchmarks, outperforming previous methods by an average of 20%. Notably, it significantly reduces hallucinations by up to 93% on ObjectHal-Bench, and also exhibits superior data efficiency towards robust and cost-effective VLM alignment.

Yubo Zhu, Dongrui Liu, Zecheng Lin et al.

Estimating the difficulty of input questions as perceived by large language models (LLMs) is essential for accurate performance evaluation and adaptive inference. Existing methods typically rely on repeated response sampling, auxiliary models, or fine-tuning the target model itself, which may incur substantial computational costs or compromise generality. In this paper, we propose a novel approach for difficulty estimation that leverages only the hidden representations produced by the target LLM. We model the token-level generation process as a Markov chain and define a value function to estimate the expected output quality given any hidden state. This allows for efficient and accurate difficulty estimation based solely on the initial hidden state, without generating any output tokens. Extensive experiments across both textual and multimodal tasks demonstrate that our method consistently outperforms existing baselines in difficulty estimation. Moreover, we apply our difficulty estimates to guide adaptive reasoning strategies, including Self-Consistency, Best-of-N, and Self-Refine, achieving higher inference efficiency with fewer generated tokens.

Zeren Chen, Xiaoya Lu, Zhijie Zheng et al.

Vision Language Models (VLMs) exhibit a fundamental semantic-to-geometric gap in spatial reasoning: they excel at qualitative semantic inference but their reasoning operates within a lossy semantic space, misaligned with high-fidelity geometry. Current paradigms fail to bridge this gap. Training-based methods suffer from an ``oracle paradox,'' learning flawed spatial logic from imperfect oracles. Tool-integrated methods constrain the final computation but critically leave the VLM's planning process unconstrained, resulting in geometrically flawed plans. In this work, we propose Geometrically-Constrained Agent (GCA), a training-free agentic paradigm that resolves this gap by introducing a formal task constraint. Specifically, we strategically decouples the VLM's role into two stages. First, acting as a semantic analyst, the VLM translates the user's ambiguous query into the formal, verifiable task constraint, which defines the reference frame and objective. Second, acting as a task solver, the VLM generates and executes tool calls strictly within the deterministic bounds defined by the constraint. This geometrically-constrained reasoning strategy successfully resolve the semantic-to-geometric gap, yielding a robust and verifiable reasoning pathway for spatial reasoning. Comprehensive experiments demonstrate that GCA achieves SOTA performance on multiple spatial reasoning benchmarks, surpassing existing training-based and tool-integrated methods by ~27%. Please see our homepage at https://gca-spatial-reasoning.github.io.

Qinghong Sun, Zhenfei Yin, Yichao Wu et al.

With the development of deep learning, the field of face anti-spoofing (FAS) has witnessed great progress. FAS is usually considered a classification problem, where each class is assumed to contain a single cluster optimized by softmax loss. In practical deployment, one class can contain several local clusters, and a single-center is insufficient to capture the inherent structure of the FAS data. However, few approaches consider large distribution discrepancies in the field of FAS. In this work, we propose a unified framework called Latent Distribution Adjusting (LDA) with properties of latent, discriminative, adaptive, generic to improve the robustness of the FAS model by adjusting complex data distribution with multiple prototypes. 1) Latent. LDA attempts to model the data of each class as a Gaussian mixture distribution, and acquire a flexible number of centers for each class in the last fully connected layer implicitly. 2) Discriminative. To enhance the intra-class compactness and inter-class discrepancy, we propose a margin-based loss for providing distribution constrains for prototype learning. 3) Adaptive. To make LDA more efficient and decrease redundant parameters, we propose Adaptive Prototype Selection (APS) by selecting the appropriate number of centers adaptively according to different distributions. 4) Generic. Furthermore, LDA can adapt to unseen distribution by utilizing very few training data without re-training. Extensive experiments demonstrate that our framework can 1) make the final representation space both intra-class compact and inter-class separable, 2) outperform the state-of-the-art methods on multiple standard FAS benchmarks.

Jing Shao, Siyu Chen, Yangguang Li et al.

Enormous waves of technological innovations over the past several years, marked by the advances in AI technologies, are profoundly reshaping the industry and the society. However, down the road, a key challenge awaits us, that is, our capability of meeting rapidly-growing scenario-specific demands is severely limited by the cost of acquiring a commensurate amount of training data. This difficult situation is in essence due to limitations of the mainstream learning paradigm: we need to train a new model for each new scenario, based on a large quantity of well-annotated data and commonly from scratch. In tackling this fundamental problem, we move beyond and develop a new learning paradigm named INTERN. By learning with supervisory signals from multiple sources in multiple stages, the model being trained will develop strong generalizability. We evaluate our model on 26 well-known datasets that cover four categories of tasks in computer vision. In most cases, our models, adapted with only 10% of the training data in the target domain, outperform the counterparts trained with the full set of data, often by a significant margin. This is an important step towards a promising prospect where such a model with general vision capability can dramatically reduce our reliance on data, thus expediting the adoption of AI technologies. Furthermore, revolving around our new paradigm, we also introduce a new data system, a new architecture, and a new benchmark, which, together, form a general vision ecosystem to support its future development in an open and inclusive manner. See project website at https://opengvlab.shlab.org.cn .

Bowen Yang, Jing Zhang, Zhenfei Yin et al.

Face anti-spoofing (FAS) is an indispensable and widely used module in face recognition systems. Although high accuracy has been achieved, a FAS system will never be perfect due to the non-stationary applied environments and the potential emergence of new types of presentation attacks in real-world applications. In practice, given a handful of labeled samples from a new deployment scenario (target domain) and abundant labeled face images in the existing source domain, the FAS system is expected to perform well in the new scenario without sacrificing the performance on the original domain. To this end, we identify and address a more practical problem: Few-Shot Domain Expansion for Face Anti-Spoofing (FSDE-FAS). This problem is challenging since with insufficient target domain training samples, the model may suffer from both overfitting to the target domain and catastrophic forgetting of the source domain. To address the problem, this paper proposes a Style transfer-based Augmentation for Semantic Alignment (SASA) framework. We propose to augment the target data by generating auxiliary samples based on photorealistic style transfer. With the assistant of the augmented data, we further propose a carefully designed mechanism to align different domains from both instance-level and distribution-level, and then stabilize the performance on the source domain with a less-forgetting constraint. Two benchmarks are proposed to simulate the FSDE-FAS scenarios, and the experimental results show that the proposed SASA method outperforms state-of-the-art methods.

Yinan He, Bei Gan, Siyu Chen et al.

The rapid progress of photorealistic synthesis techniques has reached at a critical point where the boundary between real and manipulated images starts to blur. Thus, benchmarking and advancing digital forgery analysis have become a pressing issue. However, existing face forgery datasets either have limited diversity or only support coarse-grained analysis. To counter this emerging threat, we construct the ForgeryNet dataset, an extremely large face forgery dataset with unified annotations in image- and video-level data across four tasks: 1) Image Forgery Classification, including two-way (real / fake), three-way (real / fake with identity-replaced forgery approaches / fake with identity-remained forgery approaches), and n-way (real and 15 respective forgery approaches) classification. 2) Spatial Forgery Localization, which segments the manipulated area of fake images compared to their corresponding source real images. 3) Video Forgery Classification, which re-defines the video-level forgery classification with manipulated frames in random positions. This task is important because attackers in real world are free to manipulate any target frame. and 4) Temporal Forgery Localization, to localize the temporal segments which are manipulated. ForgeryNet is by far the largest publicly available deep face forgery dataset in terms of data-scale (2.9 million images, 221,247 videos), manipulations (7 image-level approaches, 8 video-level approaches), perturbations (36 independent and more mixed perturbations) and annotations (6.3 million classification labels, 2.9 million manipulated area annotations and 221,247 temporal forgery segment labels). We perform extensive benchmarking and studies of existing face forensics methods and obtain several valuable observations.

Yuanhan Zhang, Zhenfei Yin, Jing Shao et al.

As facial interaction systems are prevalently deployed, security and reliability of these systems become a critical issue, with substantial research efforts devoted. Among them, face anti-spoofing emerges as an important area, whose objective is to identify whether a presented face is live or spoof. Recently, a large-scale face anti-spoofing dataset, CelebA-Spoof which comprised of 625,537 pictures of 10,177 subjects has been released. It is the largest face anti-spoofing dataset in terms of the numbers of the data and the subjects. This paper reports methods and results in the CelebA-Spoof Challenge 2020 on Face AntiSpoofing which employs the CelebA-Spoof dataset. The model evaluation is conducted online on the hidden test set. A total of 134 participants registered for the competition, and 19 teams made valid submissions. We will analyze the top ranked solutions and present some discussion on future work directions.

Zihao Wang, Chen Lin, Lu Sheng et al.

Recently, deep learning has been utilized to solve video recognition problem due to its prominent representation ability. Deep neural networks for video tasks is highly customized and the design of such networks requires domain experts and costly trial and error tests. Recent advance in network architecture search has boosted the image recognition performance in a large margin. However, automatic designing of video recognition network is less explored. In this study, we propose a practical solution, namely Practical Video Neural Architecture Search (PV-NAS).Our PV-NAS can efficiently search across tremendous large scale of architectures in a novel spatial-temporal network search space using the gradient based search methods. To avoid sticking into sub-optimal solutions, we propose a novel learning rate scheduler to encourage sufficient network diversity of the searched models. Extensive empirical evaluations show that the proposed PV-NAS achieves state-of-the-art performance with much fewer computational resources. 1) Within light-weight models, our PV-NAS-L achieves 78.7% and 62.5% Top-1 accuracy on Kinetics-400 and Something-Something V2, which are better than previous state-of-the-art methods (i.e., TSM) with a large margin (4.6% and 3.4% on each dataset, respectively), and 2) among median-weight models, our PV-NAS-M achieves the best performance (also a new record)in the Something-Something V2 dataset.

Kun Yuan, Quanquan Li, Jing Shao et al.

Seeking effective neural networks is a critical and practical field in deep learning. Besides designing the depth, type of convolution, normalization, and nonlinearities, the topological connectivity of neural networks is also important. Previous principles of rule-based modular design simplify the difficulty of building an effective architecture, but constrain the possible topologies in limited spaces. In this paper, we attempt to optimize the connectivity in neural networks. We propose a topological perspective to represent a network into a complete graph for analysis, where nodes carry out aggregation and transformation of features, and edges determine the flow of information. By assigning learnable parameters to the edges which reflect the magnitude of connections, the learning process can be performed in a differentiable manner. We further attach auxiliary sparsity constraint to the distribution of connectedness, which promotes the learned topology focus on critical connections. This learning process is compatible with existing networks and owns adaptability to larger search spaces and different tasks. Quantitative results of experiments reflect the learned connectivity is superior to traditional rule-based ones, such as random, residual, and complete. In addition, it obtains significant improvements in image classification and object detection without introducing excessive computation burden.

Yuanhan Zhang, Zhenfei Yin, Yidong Li et al.

As facial interaction systems are prevalently deployed, security and reliability of these systems become a critical issue, with substantial research efforts devoted. Among them, face anti-spoofing emerges as an important area, whose objective is to identify whether a presented face is live or spoof. Though promising progress has been achieved, existing works still have difficulty in handling complex spoof attacks and generalizing to real-world scenarios. The main reason is that current face anti-spoofing datasets are limited in both quantity and diversity. To overcome these obstacles, we contribute a large-scale face anti-spoofing dataset, CelebA-Spoof, with the following appealing properties: 1) Quantity: CelebA-Spoof comprises of 625,537 pictures of 10,177 subjects, significantly larger than the existing datasets. 2) Diversity: The spoof images are captured from 8 scenes (2 environments * 4 illumination conditions) with more than 10 sensors. 3) Annotation Richness: CelebA-Spoof contains 10 spoof type annotations, as well as the 40 attribute annotations inherited from the original CelebA dataset. Equipped with CelebA-Spoof, we carefully benchmark existing methods in a unified multi-task framework, Auxiliary Information Embedding Network (AENet), and reveal several valuable observations.

Yuyang Qian, Guojun Yin, Lu Sheng et al.

As realistic facial manipulation technologies have achieved remarkable progress, social concerns about potential malicious abuse of these technologies bring out an emerging research topic of face forgery detection. However, it is extremely challenging since recent advances are able to forge faces beyond the perception ability of human eyes, especially in compressed images and videos. We find that mining forgery patterns with the awareness of frequency could be a cure, as frequency provides a complementary viewpoint where either subtle forgery artifacts or compression errors could be well described. To introduce frequency into the face forgery detection, we propose a novel Frequency in Face Forgery Network (F3-Net), taking advantages of two different but complementary frequency-aware clues, 1) frequency-aware decomposed image components, and 2) local frequency statistics, to deeply mine the forgery patterns via our two-stream collaborative learning framework. We apply DCT as the applied frequency-domain transformation. Through comprehensive studies, we show that the proposed F3-Net significantly outperforms competing state-of-the-art methods on all compression qualities in the challenging FaceForensics++ dataset, especially wins a big lead upon low-quality media.

Minghua Liu, Lu Sheng, Sheng Yang et al.

3D point cloud completion, the task of inferring the complete geometric shape from a partial point cloud, has been attracting attention in the community. For acquiring high-fidelity dense point clouds and avoiding uneven distribution, blurred details, or structural loss of existing methods' results, we propose a novel approach to complete the partial point cloud in two stages. Specifically, in the first stage, the approach predicts a complete but coarse-grained point cloud with a collection of parametric surface elements. Then, in the second stage, it merges the coarse-grained prediction with the input point cloud by a novel sampling algorithm. Our method utilizes a joint loss function to guide the distribution of the points. Extensive experiments verify the effectiveness of our method and demonstrate that it outperforms the existing methods in both the Earth Mover's Distance (EMD) and the Chamfer Distance (CD).

Xihui Liu, Guojun Yin, Jing Shao et al.

Semantic image synthesis aims at generating photorealistic images from semantic layouts. Previous approaches with conditional generative adversarial networks (GAN) show state-of-the-art performance on this task, which either feed the semantic label maps as inputs to the generator, or use them to modulate the activations in normalization layers via affine transformations. We argue that convolutional kernels in the generator should be aware of the distinct semantic labels at different locations when generating images. In order to better exploit the semantic layout for the image generator, we propose to predict convolutional kernels conditioned on the semantic label map to generate the intermediate feature maps from the noise maps and eventually generate the images. Moreover, we propose a feature pyramid semantics-embedding discriminator, which is more effective in enhancing fine details and semantic alignments between the generated images and the input semantic layouts than previous multi-scale discriminators. We achieve state-of-the-art results on both quantitative metrics and subjective evaluation on various semantic segmentation datasets, demonstrating the effectiveness of our approach.

Zihao Wang, Xihui Liu, Hongsheng Li et al.

Text-image cross-modal retrieval is a challenging task in the field of language and vision. Most previous approaches independently embed images and sentences into a joint embedding space and compare their similarities. However, previous approaches rarely explore the interactions between images and sentences before calculating similarities in the joint space. Intuitively, when matching between images and sentences, human beings would alternatively attend to regions in images and words in sentences, and select the most salient information considering the interaction between both modalities. In this paper, we propose Cross-modal Adaptive Message Passing (CAMP), which adaptively controls the information flow for message passing across modalities. Our approach not only takes comprehensive and fine-grained cross-modal interactions into account, but also properly handles negative pairs and irrelevant information with an adaptive gating scheme. Moreover, instead of conventional joint embedding approaches for text-image matching, we infer the matching score based on the fused features, and propose a hardest negative binary cross-entropy loss for training. Results on COCO and Flickr30k significantly surpass state-of-the-art methods, demonstrating the effectiveness of our approach.

Guojun Yin, Bin Liu, Lu Sheng et al.

Synthesizing photo-realistic images from text descriptions is a challenging problem. Previous studies have shown remarkable progresses on visual quality of the generated images. In this paper, we consider semantics from the input text descriptions in helping render photo-realistic images. However, diverse linguistic expressions pose challenges in extracting consistent semantics even they depict the same thing. To this end, we propose a novel photo-realistic text-to-image generation model that implicitly disentangles semantics to both fulfill the high-level semantic consistency and low-level semantic diversity. To be specific, we design (1) a Siamese mechanism in the discriminator to learn consistent high-level semantics, and (2) a visual-semantic embedding strategy by semantic-conditioned batch normalization to find diverse low-level semantics. Extensive experiments and ablation studies on CUB and MS-COCO datasets demonstrate the superiority of the proposed method in comparison to state-of-the-art methods.

Guojun Yin, Lu Sheng, Bin Liu et al.

Dense captioning aims at simultaneously localizing semantic regions and describing these regions-of-interest (ROIs) with short phrases or sentences in natural language. Previous studies have shown remarkable progresses, but they are often vulnerable to the aperture problem that a caption generated by the features inside one ROI lacks contextual coherence with its surrounding context in the input image. In this work, we investigate contextual reasoning based on multi-scale message propagations from the neighboring contents to the target ROIs. To this end, we design a novel end-to-end context and attribute grounded dense captioning framework consisting of 1) a contextual visual mining module and 2) a multi-level attribute grounded description generation module. Knowing that captions often co-occur with the linguistic attributes (such as who, what and where), we also incorporate an auxiliary supervision from hierarchical linguistic attributes to augment the distinctiveness of the learned captions. Extensive experiments and ablation studies on Visual Genome dataset demonstrate the superiority of the proposed model in comparison to state-of-the-art methods.

Junting Pan, Chengyu Wang, Xu Jia et al.

This paper proposes the novel task of video generation conditioned on a SINGLE semantic label map, which provides a good balance between flexibility and quality in the generation process. Different from typical end-to-end approaches, which model both scene content and dynamics in a single step, we propose to decompose this difficult task into two sub-problems. As current image generation methods do better than video generation in terms of detail, we synthesize high quality content by only generating the first frame. Then we animate the scene based on its semantic meaning to obtain the temporally coherent video, giving us excellent results overall. We employ a cVAE for predicting optical flow as a beneficial intermediate step to generate a video sequence conditioned on the initial single frame. A semantic label map is integrated into the flow prediction module to achieve major improvements in the image-to-video generation process. Extensive experiments on the Cityscapes dataset show that our method outperforms all competing methods.

Lu Sheng, Junting Pan, Jiaming Guo et al.

Imagining multiple consecutive frames given one single snapshot is challenging, since it is difficult to simultaneously predict diverse motions from a single image and faithfully generate novel frames without visual distortions. In this work, we leverage an unsupervised variational model to learn rich motion patterns in the form of long-term bi-directional flow fields, and apply the predicted flows to generate high-quality video sequences. In contrast to the state-of-the-art approach, our method does not require external flow supervisions for learning. This is achieved through a novel module that performs bi-directional flows prediction from a single image. In addition, with the bi-directional flow consistency check, our method can handle occlusion and warping artifacts in a principled manner. Our method can be trained end-to-end based on arbitrarily sampled natural video clips, and it is able to capture multi-modal motion uncertainty and synthesizes photo-realistic novel sequences. Quantitative and qualitative evaluations over synthetic and real-world datasets demonstrate the effectiveness of the proposed approach over the state-of-the-art methods.

Xihui Liu, Zihao Wang, Jing Shao et al.

Referring expression grounding aims at locating certain objects or persons in an image with a referring expression, where the key challenge is to comprehend and align various types of information from visual and textual domain, such as visual attributes, location and interactions with surrounding regions. Although the attention mechanism has been successfully applied for cross-modal alignments, previous attention models focus on only the most dominant features of both modalities, and neglect the fact that there could be multiple comprehensive textual-visual correspondences between images and referring expressions. To tackle this issue, we design a novel cross-modal attention-guided erasing approach, where we discard the most dominant information from either textual or visual domains to generate difficult training samples online, and to drive the model to discover complementary textual-visual correspondences. Extensive experiments demonstrate the effectiveness of our proposed method, which achieves state-of-the-art performance on three referring expression grounding datasets.

Pengze Liu, Xihui Liu, Junjie Yan et al.

Pedestrian attribute recognition has attracted many attentions due to its wide applications in scene understanding and person analysis from surveillance videos. Existing methods try to use additional pose, part or viewpoint information to complement the global feature representation for attribute classification. However, these methods face difficulties in localizing the areas corresponding to different attributes. To address this problem, we propose a novel Localization Guided Network which assigns attribute-specific weights to local features based on the affinity between proposals pre-extracted proposals and attribute locations. The advantage of our model is that our local features are learned automatically for each attribute and emphasized by the interaction with global features. We demonstrate the effectiveness of our Localization Guided Network on two pedestrian attribute benchmarks (PA-100K and RAP). Our result surpasses the previous state-of-the-art in all five metrics on both datasets.

Zhao Zhong, Zichen Yang, Boyang Deng et al.

Convolutional neural networks have gained a remarkable success in computer vision. However, most usable network architectures are hand-crafted and usually require expertise and elaborate design. In this paper, we provide a block-wise network generation pipeline called BlockQNN which automatically builds high-performance networks using the Q-Learning paradigm with epsilon-greedy exploration strategy. The optimal network block is constructed by the learning agent which is trained to choose component layers sequentially. We stack the block to construct the whole auto-generated network. To accelerate the generation process, we also propose a distributed asynchronous framework and an early stop strategy. The block-wise generation brings unique advantages: (1) it yields state-of-the-art results in comparison to the hand-crafted networks on image classification, particularly, the best network generated by BlockQNN achieves 2.35% top-1 error rate on CIFAR-10. (2) it offers tremendous reduction of the search space in designing networks, spending only 3 days with 32 GPUs. A faster version can yield a comparable result with only 1 GPU in 20 hours. (3) it has strong generalizability in that the network built on CIFAR also performs well on the larger-scale dataset. The best network achieves very competitive accuracy of 82.0% top-1 and 96.0% top-5 on ImageNet.

Lu Sheng, Ziyi Lin, Jing Shao et al.

Zero-shot artistic style transfer is an important image synthesis problem aiming at transferring arbitrary style into content images. However, the trade-off between the generalization and efficiency in existing methods impedes a high quality zero-shot style transfer in real-time. In this paper, we resolve this dilemma and propose an efficient yet effective Avatar-Net that enables visually plausible multi-scale transfer for arbitrary style. The key ingredient of our method is a style decorator that makes up the content features by semantically aligned style features from an arbitrary style image, which does not only holistically match their feature distributions but also preserve detailed style patterns in the decorated features. By embedding this module into an image reconstruction network that fuses multi-scale style abstractions, the Avatar-Net renders multi-scale stylization for any style image in one feed-forward pass. We demonstrate the state-of-the-art effectiveness and efficiency of the proposed method in generating high-quality stylized images, with a series of applications include multiple style integration, video stylization and etc.

Yu Liu, Fangyin Wei, Jing Shao et al.

This paper proposes learning disentangled but complementary face features with minimal supervision by face identification. Specifically, we construct an identity Distilling and Dispelling Autoencoder (D2AE) framework that adversarially learns the identity-distilled features for identity verification and the identity-dispelled features to fool the verification system. Thanks to the design of two-stream cues, the learned disentangled features represent not only the identity or attribute but the complete input image. Comprehensive evaluations further demonstrate that the proposed features not only maintain state-of-the-art identity verification performance on LFW, but also acquire competitive discriminative power for face attribute recognition on CelebA and LFWA. Moreover, the proposed system is ready to semantically control the face generation/editing based on various identities and attributes in an unsupervised manner.

Xihui Liu, Hongsheng Li, Jing Shao et al.

The aim of image captioning is to generate captions by machine to describe image contents. Despite many efforts, generating discriminative captions for images remains non-trivial. Most traditional approaches imitate the language structure patterns, thus tend to fall into a stereotype of replicating frequent phrases or sentences and neglect unique aspects of each image. In this work, we propose an image captioning framework with a self-retrieval module as training guidance, which encourages generating discriminative captions. It brings unique advantages: (1) the self-retrieval guidance can act as a metric and an evaluator of caption discriminativeness to assure the quality of generated captions. (2) The correspondence between generated captions and images are naturally incorporated in the generation process without human annotations, and hence our approach could utilize a large amount of unlabeled images to boost captioning performance with no additional laborious annotations. We demonstrate the effectiveness of the proposed retrieval-guided method on COCO and Flickr30k captioning datasets, and show its superior captioning performance with more discriminative captions.

Xihui Liu, Haiyu Zhao, Maoqing Tian et al.

Pedestrian analysis plays a vital role in intelligent video surveillance and is a key component for security-centric computer vision systems. Despite that the convolutional neural networks are remarkable in learning discriminative features from images, the learning of comprehensive features of pedestrians for fine-grained tasks remains an open problem. In this study, we propose a new attention-based deep neural network, named as HydraPlus-Net (HP-net), that multi-directionally feeds the multi-level attention maps to different feature layers. The attentive deep features learned from the proposed HP-net bring unique advantages: (1) the model is capable of capturing multiple attentions from low-level to semantic-level, and (2) it explores the multi-scale selectiveness of attentive features to enrich the final feature representations for a pedestrian image. We demonstrate the effectiveness and generality of the proposed HP-net for pedestrian analysis on two tasks, i.e. pedestrian attribute recognition and person re-identification. Intensive experimental results have been provided to prove that the HP-net outperforms the state-of-the-art methods on various datasets.

Zhao Zhong, Junjie Yan, Wei Wu et al.

Convolutional neural networks have gained a remarkable success in computer vision. However, most usable network architectures are hand-crafted and usually require expertise and elaborate design. In this paper, we provide a block-wise network generation pipeline called BlockQNN which automatically builds high-performance networks using the Q-Learning paradigm with epsilon-greedy exploration strategy. The optimal network block is constructed by the learning agent which is trained sequentially to choose component layers. We stack the block to construct the whole auto-generated network. To accelerate the generation process, we also propose a distributed asynchronous framework and an early stop strategy. The block-wise generation brings unique advantages: (1) it performs competitive results in comparison to the hand-crafted state-of-the-art networks on image classification, additionally, the best network generated by BlockQNN achieves 3.54% top-1 error rate on CIFAR-10 which beats all existing auto-generate networks. (2) in the meanwhile, it offers tremendous reduction of the search space in designing networks which only spends 3 days with 32 GPUs, and (3) moreover, it has strong generalizability that the network built on CIFAR also performs well on a larger-scale ImageNet dataset.