Fine-tuning large language models (LLMs) improves performance but introduces critical safety vulnerabilities: even minimal harmful data can severely compromise safety measures. We observe that perturbations orthogonal to the alignment direction—defined by weight differences between aligned (safe) and unaligned models—rapidly compromise model safety. In contrast, updates along the alignment direction largely preserve it, revealing the parameter space as a "narrow safety basin". To address this, we propose AsFT (Anchoring Safety in Fine-Tuning) to maintain safety by explicitly constraining update directions during fine-tuning. By penalizing updates orthogonal to the alignment direction, AsFT effectively constrains the model within the "narrow safety basin," thus preserving its inherent safety. Extensive experiments on multiple datasets and models show that AsFT reduces harmful behaviors by up to 7.60%, improves task performance by 3.44%, and consistently outperforms existing methods across multiple tasks.

Diffusion Transformer (DiT), an emerging diffusion model for visual generation, has demonstrated superior performance but suffers from substantial computational costs. Our investigations reveal that these costs primarily stem from the static inference paradigm, which inevitably introduces redundant computation in certain diffusion timesteps and spatial regions. To overcome this inefficiency, we propose Dynamic Diffusion Transformer (DyDiT), an architecture that dynamically adjusts its computation along both timestep and spatial dimensions. Specifically, we introduce a Timestep-wise Dynamic Width (TDW) approach that adapts model width conditioned on the generation timesteps. In addition, we design a Spatial-wise Dynamic Token (SDT) strategy to avoid redundant computation at unnecessary spatial locations. TDW and SDT can be seamlessly integrated into DiT and significantly accelerate the generation process. Building on these designs, we present an extended version, DyDiT++, with improvements in three key aspects. First, it extends the generation mechanism of DyDiT beyond diffusion to flow matching, demonstrating that our method can also accelerate flow-matching-based generation, enhancing its versatility. Furthermore, we enhance DyDiT to tackle more complex visual generation tasks, including video generation and text-to-image generation, thereby broadening its real-world applications. Finally, to address the high cost of full fine-tuning and democratize technology access, we investigate the feasibility of training DyDiT in a parameter-efficient manner and introduce timestep-based dynamic LoRA (TD-LoRA). Extensive experiments on diverse visual generation models, including DiT, SiT, Latte, and FLUX, demonstrate the effectiveness of DyDiT++. Remarkably, with $< $<3% additional fine-tuning iterations, our approach reduces the FLOPs of DiT-XL by 51%, yielding 1.73× realistic speedup on hardware, and achieves a competitive FID score of 2.07 on ImageNet.

This work focuses on open-domain 4D avatarization, with the purpose of creating a 4D avatar from a portrait image in an arbitrary style. We select parametric triplanes as the intermediate 4D representation, and propose a practical training paradigm that takes advantage of both generative adversarial networks (GANs) and diffusion models. Our design stems from the observation that 4D GANs excel at bridging images and triplanes without supervision yet usually face challenges in handling diverse data distributions. A robust 2D diffusion prior emerges as the solution, assisting the GAN in transferring its expertise across various domains. The synergy between these experts permits the construction of a multi-domain image-triplane dataset, which drives the development of a general 4D avatar creator. Extensive experiments suggest that our model, termed AvatarArtist, is capable of producing high-quality 4D avatars with strong robustness to various source image domains. The code, the data, and the models will be made publicly available to facilitate future studies.

Coordinated audio generation based on video inputs typically requires a strict audio-visual (AV) alignment, where both semantics and rhythmics of the generated audio segments shall correspond to those in the video frames. Previous studies leverage a two-stage design where the AV encoders are firstly aligned via contrastive learning, then the encoded video representations guide the audio generation process. We observe that both contrastive learning and global video guidance are effective in aligning overall AV semantics while limiting temporally rhythmic synchronization. In this work, we propose Foley-Flow to first align unimodal AV encoders via masked modeling training, where the masked audio segments are recovered under the guidance of the corresponding video segments. After training, the AV encoders which are separately pretrained using only unimodal data are aligned with semantic and rhythmic consistency. Then, we develop a dynamic conditional flow for the final audio generation. Built upon the efficient velocity flow generation framework, our dynamic conditional flow utilizes temporally varying video features as the dynamic condition to guide corresponding audio segment generations. To this end, we extract coherent semantic and rhythmic representations during masked AV alignment, and use this representation of video segments to guide audio generation temporally. Our audio results are evaluated on the standard benchmarks and largely surpass existing results under several metrics. The superior performance indicates that Foley-Flow is effective in generating coordinated audios that are both semantically and rhythmically coherent to various video sequences.

Multimodal Large Language Models (MLLMs) have emerged to tackle the challenges of Visual Question Answering (VQA), sparking a new research focus on conducting objective evaluations of these models. Existing evaluation methods face limitations due to the significant human workload required to design Q&A pairs for visual images, which inherently restricts the scale and scope of evaluations. Although automated MLLM-as-judge approaches attempt to reduce the human workload through automatic evaluations, they often introduce biases. To address these problems, we propose an Unsupervised Peer review MLLM Evaluation framework. It utilizes only image data, allowing models to automatically generate questions and conduct peer review assessments of answers from other models, effectively alleviating the reliance on human workload. Additionally, we introduce the vision-language scoring system to mitigate the bias issues, which focuses on three aspects: (i) response correctness; (ii) visual understanding and reasoning; and (iii) image-text correlation. Experimental results demonstrate that UPME achieves a Pearson correlation of 0.944 with human evaluations on the MMstar dataset and 0.814 on the ScienceQA dataset, indicating that our framework closely aligns with human-designed benchmarks and inherent human preferences.

Vision-language models (VLMs) have shown remarkable success across various multi-modal tasks, yet large VLMs encounter significant efficiency challenges due to processing numerous visual tokens. A promising approach to accelerating large VLM inference is using partial information, such as attention maps from specific layers, to assess token importance and prune less essential tokens. However, our study reveals three key insights: (i) Partial attention information is insufficient for accurately identifying critical visual tokens, resulting in suboptimal performance, especially at low token retention ratios; (ii) Global attention information, such as the attention map aggregated across all layers, more effectively preserves essential tokens and maintains comparable performance under aggressive pruning. However, the attention maps from all layers require a full inference pass, which increases computational load and is therefore impractical in existing methods; and (iii) The global attention map aggregated from a small VLM closely resembles that of a large VLM, suggesting an efficient alternative. Based on these findings, we introduce a training-free method, Small VLM Guidance for accelerating Large VLMs (SGL). Specifically, we employ the attention map aggregated from a small VLM to guide visual token pruning in a large VLM. Additionally, an early exiting mechanism is developed to fully use the small VLM’s predictions, dynamically invoking the larger VLM only when necessary, yielding a superior trade-off between accuracy and computation. Extensive evaluations across 11 benchmarks demonstrate the effectiveness and generalizability of SGL, achieving up to 91% pruning ratio for visual tokens while retaining competitive performance. The code is publicly available at https://github.com/NUS-HPC-AI-Lab/SGL.

Language-based object detection (LOD) aims to align visual objects with language expressions. A large amount of paired data is utilized to improve LOD model generalizations. During the training process, recent studies leverage vision-language models (VLMs) to automatically generate human-like expressions for visual objects, facilitating training data scaling up. In this process, we observe that VLM hallucinations bring inaccurate object descriptions (e.g., object name, color, and shape) to deteriorate VL alignment quality. To reduce VLM hallucinations, we propose an agentic workflow controlled by an LLM to re-align language to visual objects via adaptively adjusting image and text prompts. We name this workflow Real-LOD, which includes planning, tool use, and reflection steps. Given an image with detected objects and VLM raw language expressions, Real-LOD reasons its state automatically and arranges action based on our neural symbolic designs (i.e., planning). The action will adaptively adjust the image and text prompts and send them to VLMs for object re-description (i.e., tool use). Then, we use another LLM to analyze these refined expressions for feedback (i.e., reflection). These steps are conducted in a cyclic form to gradually improve language descriptions for re-aligning to visual objects. We construct a dataset that contains a tiny amount of 0.18M images with re-aligned language expression and train a prevalent LOD model to surpass existing LOD methods by around 50% on the standard benchmarks. Our Real-LOD workflow, with automatic VL refinement, reveals a potential to preserve data quality along with scaling up data quantity, which further improves LOD performance from a data-alignment perspective.

Existing large language models (LLMs) evaluation methods typically focus on testing the performance on some closed-environment and domain-specific benchmarks with human annotations. In this paper, we explore a novel unsupervised evaluation direction, utilizing peer-review mechanisms to measure LLMs automatically without any human feedback. In this setting, both open-source and closed-source LLMs lie in the same environment, capable of answering unlabeled questions and evaluating each other, where each LLM’s response score is jointly determined by other anonymous ones. During this process, we found that those answers that are more recognized by other ``reviewers'' (models) usually come from LLMs with stronger abilities, while these models can also evaluate others' answers more accurately. We formalize it as a consistency assumption, i.e., the ability and score of the model usually have consistency. We exploit this to optimize each model's confidence, thereby re-ranking the LLMs to be closer to human rankings. We perform experiments on multiple datasets with standard rank-based metrics, validating the effectiveness of the proposed approach.

Diffusion Transformer (DiT), an emerging diffusion model for image generation, has demonstrated superior performance but suffers from substantial computational costs. Our investigations reveal that these costs stem from the static inference paradigm, which inevitably introduces redundant computation in certain diffusion timesteps and spatial regions. To address this inefficiency, we propose Dynamic Diffusion Transformer (DyDiT), an architecture that dynamically adjusts its compu- tation along both timestep and spatial dimensions during generation. Specifically, we introduce a Timestep-wise Dynamic Width (TDW) approach that adapts model width conditioned on the generation timesteps. In addition, we design a Spatial- wise Dynamic Token (SDT) strategy to avoid redundant computation at unnecessary spatial locations. Extensive experiments on various datasets and different-sized models verify the superiority of DyDiT. Notably, with <3% additional fine-tuning it- erations, our method reduces the FLOPs of DiT-XL by 51%, accelerates generation by 1.73×, and achieves a competitive FID score of 2.07 on ImageNet.

Training embodied agents to perform complex robotic tasks presents significant challenges due to the entangled factors of task compositionality, environmental diversity, and dynamic changes. In this work, we introduce a novel imitation learning framework to train closed-loop concept-guided policies that enhance long-horizon task performance by leveraging discovered manipulation concepts. Unlike methods that rely on predefined skills and human-annotated labels, our approach allows agents to autonomously abstract manipulation concepts from their proprioceptive states, thereby alleviating misalignment due to ambiguities in human semantics and environmental complexity. Our framework comprises two primary components: an *Automatic Concept Discovery* module that identifies meaningful and consistent manipulation concepts, and a *Concept-Guided Policy Learning* module that effectively utilizes these manipulation concepts for adaptive task execution, including a *Concept Selection Transformer* for concept-based guidance and a *Concept-Guided Policy* for action prediction with the selected concepts. Experiments demonstrate that our approach significantly outperforms baseline methods across a range of tasks and environments, while showcasing emergent consistency in motion patterns associated with the discovered manipulation concepts. Codes are available at: https://github.com/PeiZhou26/AutoCGP.

Model merging has emerged as a powerful technique for combining task-specific vision models into a unified and multi-functional model. Previous methods represented by task arithmetic, have demonstrated effectiveness and scalability in this domain. When large vision-language models (LVLMs) arise with model size scaling up, this design becomes challenging to fuse different instruction-tuned LVLMs for generalization enhancement. The large scale and multi-modal nature of LVLMs present unique obstacles, including constructing reusable and modular components to accommodate the multi-component architecture of LVLMs and the requirement for dynamic fusion based on multi-modal input tokens. To address these challenges, we propose the \textbf{RE}cipe \textbf{ME}rging \textbf{DY}namics (REMEDY) method, a scalable and flexible paradigm for model merging in LVLMs. We first define reusable modules termed \textit{recipes} including the projector and shallow LLM layers, enhancing visual-language understanding. Then, we introduce a modality-aware allocator dynamically generates weights in a one-shot manner based on input relevance to existing recipes, enabling efficient cross-modal knowledge integration. REMEDY thus offers an adaptive solution for LVLMs to tackle both seen (i.e., multi-task learning) and unseen (i.e., zero-shot generalization) tasks. Experimental results demonstrate that our method consistently improves performance on both seen and unseen tasks, underscoring the effectiveness of REMEDY in diverse multi-modal scenarios.

Current text-to-image (T2I) generation models struggle to align spatial composition with the input text, especially in complex scenes. Even layout-based approaches yield suboptimal spatial control, as their generation process is decoupled from layout planning, making it difficult to refine the layout during synthesis. We present CoT-Diff, a framework that brings step-by-step CoT-style reasoning into T2I generation by tightly integrating Multimodal Large Language Model (MLLM)-driven 3D layout planning with the diffusion process. CoT-Diff enables layout-aware reasoning inline within a single diffusion round: at each denoising step, the MLLM evaluates intermediate predictions, dynamically updates the 3D scene layout, and continuously guides the generation process. The updated layout is converted into semantic conditions and depth maps, which are fused into the diffusion model via a condition-aware attention mechanism, enabling precise spatial control and semantic injection. Experiments on 3D Scene benchmarks show that CoT-Diff significantly improves spatial alignment and compositional fidelity, and outperforms the state-of-the-art method by 34.7% in complex scene spatial accuracy, thereby validating the effectiveness of this entangled generation paradigm.

The relation modeling between actors and scene context advances video action detection where the correlation of multiple actors makes their action recognition challenging. Existing studies model each actor and scene relation to improve action recognition. However, the scene variations and background interference limit their effectiveness. In this paper, we propose to select actor-related scene context, rather than directly laveraging raw video scenario, to improve relation modeling. We develop a Cycle Actor-Context Relation network (CycleACR) where there is a symmetric graph that models the actor and context relations in a bidirectional form. Specifically, our CycleACR is constituted of two modules: 1) Actor-to-Context Reorganization (A2C-R), which adaptively collects actor features for context feature reorganizations, and 2) Context-to-Actor Enhancement (C2A-E), which dynamically utilizes the reorganized context features for actor feature enhancement. Stacking multiple CycleACR modules is able to effectively capture the high-order relation and efficiently exchange useful information between actors and context. To fully exploit time-dependent and holistic context information, we further design a parallel local and global temporal context modeling branch. The outputs of the two branches are integrated as the final context-enhanced actor feature representations. Finally, we propose a context-aware memory bank for long-term relation modeling. The proposed bank can effectively store actor-related scene context from other clips without additional memory overhead. Compared to existing designs that focus on C2A-E, our CycleACR introduces the core design of A2C-R for more effective relation modeling. This cycle modeling enablesour CycleACR to achieve state-of-the-art performance on two popular action detection datasets: AVA (40.6 mAP) and UCF101-24 (84.7 mAP). We also provide ablation studies and visualizations to show how our cycle actor-context relation modeling improves video action detection.

Generative adversarial networks (GANs) have made great success in image inpainting yet still have difficulties tackling large missing regions. In contrast, iterative probabilistic algorithms, such as autoregressive and denoising diffusion models, have to be deployed with massive computing resources for decent effect. To achieve high-quality results with low computational cost, we present a novel pixel spread model (PSM) that iteratively employs decoupled probabilistic modeling, combining the optimization efficiency of GANs with the prediction tractability of probabilistic models. As a result, our model selectively spreads informative pixels throughout the image in a few iterations, largely enhancing the completion quality and efficiency. On multiple benchmarks, we achieve new state-of-the-art performance. Our code and models will be publicly available.

We propose InstructDET, a data-centric method for referring object detection (ROD) that localizes target objects based on user instructions. While deriving from referring expressions (REC), the instructions we leverage are greatly diversified to encompass common user intentions related to object detection. For one image, we produce tremendous instructions that refer to every single object and different combinations of multiple objects. Each instruction and its corresponding object bounding boxes (bbxs) constitute one training data pair. In order to encompass common detection expressions, we involve emerging vision-language model (VLM) and large language model (LLM) to generate instructions guided by text prompts and object bbxs, as the generalizations of foundation models are effective to produce human-like expressions (e.g., describing object property, category, and relationship). We name our constructed dataset as InDET. It contains images, bbxs and generalized instructions that are from foundation models. Our InDET is developed from existing REC datasets and object detection datasets, with the expanding potential that any image with object bbxs can be incorporated through using our InstructDET method. By using our InDET dataset, we show that a conventional ROD model surpasses existing methods on standard REC datasets and our InDET test set. Our data-centric method InstructDET, with automatic data expansion by leveraging foundation models, directs a promising field that ROD can be greatly diversified to execute common object detection instructions.

Domain generalization (DG) methods aim to maintain good performance in an unseen target domain by using training data from multiple source domains. While success on certain occasions are observed, enhancing the baseline across most scenarios remains challenging. This work introduces a simple yet effective framework, dubbed learning from multiple experts (LFME), that aims to make the target model an expert in all source domains to improve DG. Specifically, besides learning the target model used in inference, LFME will also train multiple experts specialized in different domains, whose output probabilities provide professional guidance by simply regularizing the logit of the target model. Delving deep into the framework, we reveal that the introduced logit regularization term implicitly provides effects of enabling the target model to harness more information, and mining hard samples from the experts during training. Extensive experiments on benchmarks from different DG tasks demonstrate that LFME is consistently beneficial to the baseline and can achieve comparable performance to existing arts. Code is available at https://github.com/liangchen527/LFME.

Visual content and accompanied audio signals naturally formulate a joint representation to improve audio-visual (AV) related applications. While studies develop various AV representation learning frameworks, the importance of AV data alignment is usually undermined for achieving high-quality representation. We observe that an audio signal may contain background noise interference. Also, non-synchronization may appear between audio and video streams. These non-strict data alignment limits representation quality and downgrade application performance. In this paper, we propose to improve AV joint representations from a data-centric perspective by aligning audio signals to visual data. Our alignment is conducted in an agentic workflow controlled by an LLM-based assistant named AVAgent. For each input AV data pair, our AVAgent uses a multi-modal LLM to convert audio and visual data into language descriptions separately (i.e., tool use). Then, AVAgent reasons whether this paired data is aligned well and plans to edit the audio signal if needed (i.e., planning). The audio editing is executed by predefined actions that filter noise or augment data. Moreover, we use a VLM to evaluate how modified audio signals match the visual content and provide feedback to AVAgent (i.e., reflection). The tool use, planning, and reflection steps operate cyclically to become an agentic workflow where audio signals are gradually aligned to visual content. To this end, existing methods can directly leverage the aligned AV data via our agentic workflow to improve AV joint representations. The experimental results comprehensively demonstrate the state-of-the-art performance of the proposed approach against previous baselines in diverse downstream tasks.

Existing parameter-efficient fine-tuning (PEFT) methods have achieved significant success on vision transformers (ViTs) adaptation by improving parameter efficiency. However, the exploration of enhancing inference efficiency during adaptation remains underexplored. This limits the broader application of pre-trained ViT models, especially when the model is computationally extensive. In this paper, we propose Dynamic Tuning (DyT), a novel approach to improve both parameter and inference efficiency for ViT adaptation. Specifically, besides using the lightweight adapter modules, we propose a token dispatcher to distinguish informative tokens from less important ones, allowing the latter to dynamically skip the original block, thereby reducing the redundant computation during inference. Additionally, we explore multiple design variants to find the best practice of DyT. Finally, inspired by the mixture-of-experts (MoE) mechanism, we introduce an enhanced adapter to further boost the adaptation performance. We validate DyT across various tasks, including image/video recognition and semantic segmentation. For instance, DyT achieves superior performance compared to existing PEFT methods while evoking only 71% of their FLOPs on the VTAB-1K benchmark.

Visual attention advances object detection by attending neural networks to object representations. While existing methods incorporate empirical modules to empower network attention, we rethink attentive object detection from the network learning perspective in this work. We propose a NEural Attention Learning approach (NEAL) which consists of two parts. During the back-propagation of each training iteration, we first calculate the partial derivatives (a.k.a. the accumulated gradients) of the classification output with respect to the input features. We refine these partial derivatives to obtain attention response maps whose elements reflect the contributions to the final network predictions. Then, we formulate the attention response maps as extra objective functions, which are combined together with the original detection loss to train detectors in an end-to-end manner. In this way, we succeed in learning an attentive CNN model without introducing additional network structures. We apply NEAL to the two-stage object detection frameworks, which are usually composed of a CNN feature backbone, a region proposal network (RPN), and a classifier. We show that the proposed NEAL not only helps the RPN attend to objects but also enables the classifier to pay more attention to the premier positive samples. To this end, the localization (proposal generation) and classification mutually benefit from each other in our proposed method. Extensive experiments on large-scale benchmark datasets, including MS COCO 2017 and Pascal VOC 2012, demonstrate that the proposed NEAL algorithm advances the two-stage object detector over state-of-the-art approaches.

The main challenge in domain generalization (DG) is to handle the distribution shift problem that lies between the training and test data. Recent studies suggest that test-time training (TTT), which adapts the learned model with test data, might be a promising solution to the problem. Generally, a TTT strategy hinges its performance on two main factors: selecting an appropriate auxiliary TTT task for updating and identifying reliable parameters to update during the test phase. Both previous arts and our experiments indicate that TTT may not improve but be detrimental to the learned model if those two factors are not properly considered. This work addresses those two factors by proposing an Improved Test-Time Adaptation (ITTA) method. First, instead of heuristically defining an auxiliary objective, we propose a learnable consistency loss for the TTT task, which contains learnable parameters that can be adjusted toward better alignment between our TTT task and the main prediction task. Second, we introduce additional adaptive parameters for the trained model, and we suggest only updating the adaptive parameters during the test phase. Through extensive experiments, we show that the proposed two strategies are beneficial for the learned model (see Figure 1), and ITTA could achieve superior performance to the current state-of-the-art methods on several DG benchmarks. Code is available at https://github.com/liangchen527/ITTA.

Visual grounding (VG) aims to establish fine-grained alignment between vision and language. Ideally, it can be a testbed for vision-and-language models to evaluate their understanding of the images and texts and their reasoning abilities over their joint space. However, most existing VG datasets are constructed using simple description texts, which do not require sufficient reasoning over the images and texts. This has been demonstrated in a recent study [27], where a simple LSTM-based text encoder without pretraining can achieve state-of-the-art performance on mainstream VG datasets. Therefore, in this paper, we propose a novel benchmark of Scene Knowledge-guided Visual Grounding (SK-VG), where the image content and referring expressions are not sufficient to ground the target objects, forcing the models to have a reasoning ability on the long-form scene knowledge. To perform this task, we propose two approaches to accept the triple-type input, where the former embeds knowledge into the image features before the image-query interaction; the latter leverages linguistic structure to assist in computing the image-text matching. We conduct extensive experiments to analyze the above methods and show that the proposed approaches achieve promising results but still leave room for improvement, including performance and interpretability. The dataset and code are available at https://github.com/zhjohnchan/SK-VG.

GAN inversion and editing via StyleGAN maps an input image into the embedding spaces (W, W+, and F) to simultaneously maintain image fidelity and meaningful manipulation. From latent space W to extended latent space W+ to feature space F in StyleGAN, the editability of GAN inversion decreases while its reconstruction quality increases. Recent GAN inversion methods typically explore W+ and F rather than W to improve reconstruction fidelity while maintaining editability. As W+ and F are derived from W that is essentially the foundation latent space of StyleGAN, these GAN inversion methods focusing on W+ and F spaces could be improved by stepping back to W. In this work, we propose to first obtain the proper latent code in foundation latent space W. We introduce contrastive learning to align W and the image space for proper latent code discovery. Then, we leverage a cross-attention encoder to transform the obtained latent code in W into W+ and F, accordingly. Our experiments show that our exploration of the foundation latent space W improves the representation ability of latent codes in W+ and features in F, which yields state-of-the-art reconstruction fidelity and editability results on the standard benchmarks. Project page: https://kumapowerliu.github.io/CLCAE.

We propose DiffusionDet, a new framework that formulates object detection as a denoising diffusion process from noisy boxes to object boxes. During the training stage, object boxes diffuse from ground-truth boxes to random distribution, and the model learns to reverse this noising process. In inference, the model refines a set of randomly generated boxes to the output results in a progressive way. Our work possesses an appealing property of flexibility, which enables the dynamic number of boxes and iterative evaluation. The extensive experiments on the standard benchmarks show that DiffusionDet achieves favorable performance compared to previous well-established detectors. For example, DiffusionDet achieves 5.3 AP and 4.8 AP gains when evaluated with more boxes and iteration steps, under a zero-shot transfer setting from COCO to CrowdHuman. Our code is available at https://github.com/ShoufaChen/DiffusionDet.

Streaming video clips with large-scale video tokens impede vision transformers (ViTs) for efficient recognition, especially in video action detection where sufficient spatiotemporal representations are required for precise actor identification. In this work, we propose an end-to-end framework for efficient video action detection (EVAD) based on vanilla ViTs. Our EVAD consists of two specialized designs for video action detection. First, we propose a spatiotemporal token dropout from a keyframe-centric perspective. In a video clip, we maintain all tokens from its keyframe, preserve tokens relevant to actor motions from other frames, and drop out the remaining tokens in this clip. Second, we refine scene context by leveraging remaining tokens for better recognizing actor identities. The region of interest (RoI) in our action detector is expanded into temporal domain. The captured spatiotemporal actor identity representations are refined via scene context in a decoder with the attention mechanism. These two designs make our EVAD efficient while maintaining accuracy, which is validated on three benchmark datasets (i.e., AVA, UCF101-24, JHMDB). Compared to the vanilla ViT backbone, our EVAD reduces the overall GFLOPs by 43% and improves real-time inference speed by 40% with no performance degradation. Moreover, even at similar computational costs, our EVAD can improve the performance by 1.1 mAP with higher resolution inputs. Code is available at https://github.com/MCG-NJU/EVAD.

This paper offers a new perspective to ease the challenge of domain generalization, which involves maintaining robust results even in unseen environments. Our design focuses on the decision-making process in the final classifier layer. Specifically, we propose treating the element-wise contributions to the final results as the rationale for making a decision and representing the rationale for each sample as a matrix. For a well-generalized model, we suggest the rationale matrices for samples belonging to the same category should be similar, indicating the model relies on domain-invariant clues to make decisions, thereby ensuring robust results. To implement this idea, we introduce a rationale invariance loss as a simple regularization technique, requiring only a few lines of code. Our experiments demonstrate that the proposed approach achieves competitive results across various datasets, despite its simplicity. Code is available at https://github.com/liangchen527/RIDG.

Parameter Efficient Tuning (PET) has gained attention for reducing the number of parameters while maintaining performance and providing better hardware resource savings, but few studies investigate dense prediction tasks and interaction between modalities. In this paper, we do an investigation of efficient tuning problems on referring image segmentation. We propose a novel adapter called Bridger to facilitate cross-modal information exchange and inject task-specific information into the pre-trained model. We also design a lightweight decoder for image segmentation. Our approach achieves comparable or superior performance with only 1.61% to 3.38% backbone parameter updates, evaluated on challenging benchmarks. The code is available at https://github.com/kkakkkka/ETRIS.

Although considerable progress has been made in image deraining under synthetic data, real rain removal is still a tough problem due to the huge domain gap between synthetic and real data. Besides, difficulties in collecting and labeling diverse real rain images hinder the progress of this field. Consequently, we attempt to promote real rain removal from rain image generation (RIG) perspective. Existing RIG methods mainly focus on diversity but miss realistic, or the realistic but neglect diversity of the generation. To solve this dilemma, we propose a physical alignment and controllable generation network (PCGNet) for diverse and realistic rain generation. Our key idea is to simultaneously utilize the controllability of attributes from synthetic and the realism of appearance from real data. Specifically, we devise a unified framework to disentangle background, rain attributes, and appearance style from synthetic and real data. Then we collaboratively align the factors with a novel semi-supervised weight moving strategy for attribute, an explicit distribution modeling method for real rain style. Furthermore, we pack these aligned factors into the generation model, achieving physical controllable mapping from the attributes to real rain with image-level and attribute-level consistency loss. Extensive experiments show that PCGNet can effectively generate appealing rainy results, which significantly improve the performance under synthetic and real scenes for all existing deraining methods.

In zero-shot learning (ZSL), generative methods synthesize class-related sample features based on predefined semantic prototypes. They advance the ZSL performance by synthesizing unseen class sample features for better training the classifier. We observe that each class’s predefined semantic prototype (also referred to as semantic embedding or condition) does not accurately match its real semantic prototype. So the synthesized visual sample features do not faithfully represent the real sample features, limiting the classifier training and existing ZSL performance. In this paper, we formulate this mismatch phenomenon as the visual-semantic domain shift problem. We propose a dynamic semantic prototype evolving (DSP) method to align the empirically predefined semantic prototypes and the real prototypes for class-related feature synthesis. The alignment is learned by refining sample features and semantic prototypes in a unified framework and making the synthesized visual sample features approach real sample features. After alignment, synthesized sample features from unseen classes are closer to the real sample features and benefit DSP to improve existing generative ZSL methods by 8.5%, 8.0%, and 9.7% on the standard CUB, SUN AWA2 datasets, the significant performance improvement indicates that evolving semantic prototype explores a virgin field in ZSL.

Zero-shot learning (ZSL) tackles the novel class recognition problem by transferring semantic knowledge from seen classes to unseen ones. Semantic knowledge is typically represented by attribute descriptions shared between different classes, which act as strong priors for localizing object attributes that represent discriminative region features, enabling significant and sufficient visual-semantic interaction for advancing ZSL. Existing attention-based models have struggled to learn inferior region features in a single image by solely using unidirectional attention, which ignore the transferable and discriminative attribute localization of visual features for representing the key semantic knowledge for effective knowledge transfer in ZSL. In this paper, we propose a cross attribute-guided Transformer network, termed TransZero++, to refine visual features and learn accurate attribute localization for key semantic knowledge representations in ZSL. Specifically, TransZero++ employs an attribute → visual Transformer sub-net (AVT) and a visual → attribute Transformer sub-net (VAT) to learn attribute-based visual features and visual-based attribute features, respectively. By further introducing feature-level and prediction-level semantical collaborative losses, the two attribute-guided transformers teach each other to learn semantic-augmented visual embeddings for key semantic knowledge representations via semantical collaborative learning. Finally, the semantic-augmented visual embeddings learned by AVT and VAT are fused to conduct desirable visual-semantic interaction cooperated with class semantic vectors for ZSL classification. Extensive experiments show that TransZero++ achieves the new state-of-the-art results on three golden ZSL benchmarks and on the large-scale ImageNet dataset. The project website is available at: https://shiming-chen.github.io/TransZero-pp/TransZero-pp.html.

Recent studies in deepfake detection have yielded promising results when the training and testing face forgeries are from the same dataset. However, the problem remains challenging when one tries to generalize the detector to forgeries created by unseen methods in the training dataset. This work addresses the generalizable deepfake detection from a simple principle: a generalizable representation should be sensitive to diverse types of forgeries. Following this principle, we propose to enrich the “diversity” of forgeries by synthesizing augmented forgeries with a pool of forgery configurations and strengthen the “sensitivity” to the forgeries by enforcing the model to predict the forgery configurations. To effectively explore the large forgery augmentation space, we further propose to use the adversarial training strategy to dynamically synthesize the most challenging forgeries to the current model. Through extensive experiments, we show that the proposed strategies are surprisingly effective (see Figure 1), and they could achieve superior performance than the current state-of-the-art methods. Code is available at https://github.com/liangchen527/SLADD.

Recently, MLP-like vision models have achieved promising performances on mainstream visual recognition tasks. In contrast with vision transformers and CNNs, the success of MLP-like models shows that simple information fusion operations among tokens and channels can yield a good representation power for deep recognition models. However, existing MLP-like models fuse tokens through static fusion operations, lacking adaptability to the contents of the tokens to be mixed. Thus, customary information fusion procedures are not effective enough. To this end, this paper presents an efficient MLP-like network architecture, dubbed DynaMixer, resorting to dynamic information fusion. Critically, we propose a procedure, on which the DynaMixer model relies, to dynamically generate mixing matrices by leveraging the contents of all the tokens to be mixed. To reduce the time complexity and improve the robustness, a dimensionality reduction technique and a multi-segment fusion mechanism are adopted. Our proposed DynaMixer model (97M parameters) achieves 84.3% top-1 accuracy on the ImageNet-1K dataset without extra training data, performing favorably against the state-of-the-art vision MLP models. When the number of parameters is reduced to 26M, it still achieves 82.7% top-1 accuracy, surpassing the existing MLP-like models with a similar capacity. The code is available at \url{https://github.com/ziyuwwang/DynaMixer}.

Pretraining Vision Transformers (ViTs) has achieved great success in visual recognition. A following scenario is to adapt a ViT to various image and video recognition tasks. The adaptation is challenging because of heavy computation and memory storage. Each model needs an independent and complete finetuning process to adapt to different tasks, which limits its transferability to different visual domains.To address this challenge, we propose an effective adaptation approach for Transformer, namely AdaptFormer, which can adapt the pre-trained ViTs into many different image and video tasks efficiently.It possesses several benefits more appealing than prior arts.Firstly, AdaptFormer introduces lightweight modules that only add less than 2% extra parameters to a ViT, while it is able to increase the ViT's transferability without updating its original pre-trained parameters, significantly outperforming the existing 100\% fully fine-tuned models on action recognition benchmarks.Secondly, it can be plug-and-play in different Transformers and scalable to many visual tasks.Thirdly, extensive experiments on five image and video datasets show that AdaptFormer largely improves ViTs in the target domains. For example, when updating just 1.5% extra parameters, it achieves about 10% and 19% relative improvement compared to the fully fine-tuned models on Something-Something~v2 and HMDB51, respectively. Code is available at https://github.com/ShoufaChen/AdaptFormer.

State-of-the-art deepfake detectors perform well in identifying forgeries when they are evaluated on a test set similar to the training set, but struggle to maintain good performance when the test forgeries exhibit different characteristics from the training images e.g., forgeries are created by unseen deepfake methods. Such a weak generalization capability hinders the applicability of deepfake detectors. In this paper, we introduce a new learning paradigm specially designed for the generalizable deepfake detection task. Our key idea is to construct a test-sample-specific auxiliary task to update the model before applying it to the sample. Specifically, we synthesize pseudo-training samples from each test image and create a test-time training objective to update the model. Moreover, we proposed to leverage meta-learning to ensure that a fast single-step test-time gradient descent, dubbed one-shot test-time training (OST), can be sufficient for good deepfake detection performance. Extensive results across several benchmark datasets demonstrate that our approach performs favorably against existing arts in terms of generalization to unseen data and robustness to different post-processing steps.

Pre-training video transformers on extra large-scale datasets is generally required to achieve premier performance on relatively small datasets. In this paper, we show that video masked autoencoders (VideoMAE) are data-efficient learners for self-supervised video pre-training (SSVP). We are inspired by the recent ImageMAE and propose customized video tube masking with an extremely high ratio. This simple design makes video reconstruction a more challenging and meaningful self-supervision task, thus encouraging extracting more effective video representations during the pre-training process. We obtain three important findings with VideoMAE: (1) An extremely high proportion of masking ratio (i.e., 90% to 95%) still yields favorable performance for VideoMAE. The temporally redundant video content enables higher masking ratio than that of images. (2) VideoMAE achieves impressive results on very small datasets (i.e., around 3k-4k videos) without using any extra data. This is partially ascribed to the challenging task of video reconstruction to enforce high-level structure learning. (3) VideoMAE shows that data quality is more important than data quantity for SSVP. Domain shift between pre-training and target datasets is an important factor. Notably, our VideoMAE with the vanilla ViT backbone can achieve 87.4% on Kinects-400, 75.4% on Something-Something V2, 91.3% on UCF101, and 62.6% on HMDB51, without using any extra data. Code is available at https://github.com/MCG-NJU/VideoMAE.

Free-form text prompts allow users to describe their intentions during image manipulation conveniently. Based on the visual latent space of StyleGAN[21] and text embedding space of CLIP[34], studies focus on how to map these two latent spaces for text-driven attribute manipulations. Currently, the latent mapping between these two spaces is empirically designed and confines that each manipulation model can only handle one fixed text prompt. In this paper, we propose a method named Free-Form CLIP (FFCLIP), aiming to establish an automatic latent mapping so that one manipulation model handles free-form text prompts. Our FFCLIP has a cross-modality semantic modulation module containing semantic alignment and injection. The semantic alignment performs the automatic latent mapping via linear transformations with a cross attention mechanism. After alignment, we inject semantics from text prompt embeddings to the StyleGAN latent space. For one type of image (e.g., human portrait'), one FFCLIP model can be learned to handle free-form text prompts. Meanwhile, we observe that although each training text prompt only contains a single semantic meaning, FFCLIP can leverage text prompts with multiple semantic meanings for image manipulation. In the experiments, we evaluate FFCLIP on three types of images (i.e.,human portraits', cars', andchurches'). Both visual and numerical results show that FFCLIP effectively produces semantically accurate and visually realistic images. Project page: https://github.com/KumapowerLIU/FFCLIP.

MoCo [11] is effective for unsupervised image representation learning. In this paper, we propose VideoMoCo for unsupervised video representation learning. Given a video sequence as an input sample, we improve the temporal feature representations of MoCo from two perspectives. First, we introduce a generator to drop out several frames from this sample temporally. The discriminator is then learned to encode similar feature representations regardless of frame removals. By adaptively dropping out different frames during training iterations of adversarial learning, we augment this input sample to train a tempo-rally robust encoder. Second, we use temporal decay to model key attenuation in the memory queue when computing the contrastive loss. As the momentum encoder updates after keys enqueue, the representation ability of these keys degrades when we use the current input sample for contrastive learning. This degradation is reflected via temporal decay to attend the input sample to recent keys in the queue. As a result, we adapt MoCo to learn video representations without empirically designing pretext tasks. By empowering the temporal robustness of the encoder and modeling the temporal decay of the keys, our VideoMoCo improves MoCo temporally based on contrastive learning. Experiments on benchmark datasets including UCF101 and HMDB51 show that VideoMoCo stands as a state-of-the-art video representation learning method.

Universal style transfer retains styles from reference images in content images. While existing methods have achieved state-of-the-art style transfer performance, they are not aware of the content leak phenomenon that the image content may corrupt after several rounds of stylization process. In this paper, we propose ArtFlow to prevent content leak during universal style transfer. ArtFlow consists of reversible neural flows and an unbiased feature transfer module. It supports both forward and backward inferences and operates in a projection-transfer-reversion scheme. The forward inference projects input images into deep features, while the backward inference remaps deep features back to input images in a lossless and unbiased way. Extensive experiments demonstrate that ArtFlow achieves comparable performance to state-of-the-art style transfer methods while avoiding content leak.

Image virtual try-on replaces the clothes on a person image with a desired in-shop clothes image. It is challenging because the person and the in-shop clothes are unpaired. Existing methods formulate virtual try-on as either in-painting or cycle consistency. Both of these two formulations encourage the generation networks to reconstruct the input image in a self-supervised manner. However, existing methods do not differentiate clothing and non-clothing regions. A straightforward generation impedes the virtual try-on quality because of the heavily coupled image contents. In this paper, we propose a Disentangled Cycle-consistency Try-On Network (DCTON). The DCTON is able to produce highly-realistic try-on images by disentangling important components of virtual try-on including clothes warping, skin synthesis, and image composition. Moreover, DCTON can be naturally trained in a self-supervised manner following cycle consistency learning. Extensive experiments on challenging benchmarks show that DCTON outperforms state-of-the-art approaches favorably.

Image virtual try-on aims to fit a garment image (target clothes) to a person image. Prior methods are heavily based on human parsing. However, slightly-wrong segmentation results would lead to unrealistic try-on images with large artifacts. A recent pioneering work employed knowledge distillation to reduce the dependency of human parsing, where the try-on images produced by a parser-based method are used as supervisions to train a "student" network without relying on segmentation, making the student mimic the try-on ability of the parser-based model. However, the image quality of the student is bounded by the parser-based model. To address this problem, we propose a novel approach, "teacher-tutor-student" knowledge distillation, which is able to produce highly photo-realistic images without human parsing, possessing several appealing advantages compared to prior arts. (1) Unlike existing work, our approach treats the fake images produced by the parser-based method as "tutor knowledge", where the artifacts can be corrected by real "teacher knowledge", which is extracted from the real person images in a self-supervised way. (2) Other than using real images as supervisions, we formulate knowledge distillation in the try-on problem as distilling the appearance flows between the person image and the garment image, enabling us to find accurate dense correspondences between them to produce high-quality results. (3) Extensive evaluations show large superiority of our method (see Fig. 1).

Adversarial attack arises due to the vulnerability of deep neural networks to perceive input samples injected with imperceptible perturbations. Recently, adversarial attack has been applied to visual object tracking to evaluate the robustness of deep trackers. Assuming that the model structures of deep trackers are known, a variety of white-box attack approaches to visual tracking have demonstrated promising results. However, the model knowledge about deep trackers is usually unavailable in real applications. In this paper, we propose a decision-based black-box attack method for visual object tracking. In contrast to existing black-box adversarial attack methods that deal with static images for image classification, we propose IoU attack that sequentially generates perturbations based on the predicted IoU scores from both current and historical frames. By decreasing the IoU scores, the proposed attack method degrades the accuracy of temporal coherent bounding boxes (i.e., object motions) accordingly. In addition, we transfer the learned perturbations to the next few frames to initialize temporal motion attack. We validate the proposed IoU attack on state-of-the-art deep trackers (i.e., detection based, correlation filter based, and long-term trackers). Extensive experiments on the benchmark datasets indicate the effectiveness of the proposed IoU attack method. The source code is available at https://github.com/VISION-SJTU/IoUattack.

User-intended visual content fills the hole regions of an input image in the image editing scenario. The coarse low- level inputs, which typically consist of sparse sketch lines and color dots, convey user intentions for content creation (i.e., free-form editing). While existing methods combine an input image and these low-level controls for CNN inputs, the corresponding feature representations are not sufficient to convey user intentions, leading to unfaithfully generated content. In this paper, we propose DeFLOCNet which relies on a deep encoder-decoder CNN to retain the guidance of these controls in the deep feature representations. In each skip-connection layer, we design a structure generation block. Instead of attaching low-level controls to an input image, we inject these controls directly into each structure generation block for sketch line refinement and color propagation in the CNN feature space. We then concatenate the modulated features with the original decoder features for structure generation. Meanwhile, DeFLOCNet involves another decoder branch for texture generation and detail enhancement. Both structures and textures are rendered in the decoder, leading to user-intended editing results. Experiments on benchmarks demonstrate that DeFLOCNet effectively transforms different user intentions to create visually pleasing content.

We propose PD-GAN, a probabilistic diverse GAN for image inpainting. Given an input image with arbitrary hole regions, PD-GAN produces multiple inpainting results with diverse and visually realistic content. Our PD-GAN is built upon a vanilla GAN which generates images based on random noise. During image generation, we modulate deep features of input random noise from coarse-to-fine by injecting an initially restored image and the hole regions in multiple scales. We argue that during hole filling, the pixels near the hole boundary should be more deterministic (i.e., with higher probability trusting the context and initially restored image to create natural inpainting boundary), while those pixels lie in the center of the hole should enjoy more degrees of freedom (i.e., more likely to depend on the random noise for enhancing diversity). To this end, we propose spatially probabilistic diversity normalization (SPDNorm) inside the modulation to model the probability of generating a pixel conditioned on the context information. SPDNorm dynamically balances the realism and diversity inside the hole region, making the generated content more diverse towards the hole center and resemble neighboring image content more towards the hole boundary. Meanwhile, we propose a perceptual diversity loss to further empower PD-GAN for diverse content generation. Experiments on benchmark datasets including CelebA-HQ, Places2 and Paris Street View indicate that PD-GAN is effective for diverse and visually realistic image restoration.

Studies on self-supervised visual representation learning (SSL) improve encoder backbones to discriminate training samples without labels. While CNN encoders via SSL achieve comparable recognition performance to those via supervised learning, their network attention is under-explored for further improvement. Motivated by the transformers that explore visual attention effectively in recognition scenarios, we propose a CNN Attention REvitalization (CARE) framework to train attentive CNN encoders guided by transformers in SSL. The proposed CARE framework consists of a CNN stream (C-stream) and a transformer stream (T-stream), where each stream contains two branches. C-stream follows an existing SSL framework with two CNN encoders, two projectors, and a predictor. T-stream contains two transformers, two projectors, and a predictor. T-stream connects to CNN encoders and is in parallel to the remaining C-Stream. During training, we perform SSL in both streams simultaneously and use the T-stream output to supervise C-stream. The features from CNN encoders are modulated in T-stream for visual attention enhancement and become suitable for the SSL scenario. We use these modulated features to supervise C-stream for learning attentive CNN encoders. To this end, we revitalize CNN attention by using transformers as guidance. Experiments on several standard visual recognition benchmarks, including image classification, object detection, and semantic segmentation, show that the proposed CARE framework improves CNN encoder backbones to the state-of-the-art performance.

Facilitated by deep neural networks, numerous tracking methods have made significant advances. Existing deep trackers mainly utilize independent frames to model the target appearance, while paying less attention to its temporal coherence. In this paper, we propose a recurrent memory activation network (RMAN) to exploit the untapped temporal coherence of the target appearance for visual tracking. We build the RMAN on top of the long short-term memory network (LSTM) with an additional memory activation layer. Specifically, we first use the LSTM to model the temporal changes of the target appearance. Then we selectively activate the memory blocks via the activation layer to produce a temporally coherent representation. The recurrent memory activation layer enriches the target representations from independent frames and reduces the background interference through temporal consistency. The proposed RMAN is fully differentiable and can be optimized end-to-end. To facilitate network training, we propose a temporal coherence loss together with the original binary classification loss. Extensive experimental results on standard benchmarks demonstrate that our method performs favorably against the state-of-the-art approaches.

Siamese networks are prevalent in visual tracking because of the efficient localization. The networks take both a search patch and a target template as inputs where the target template is usually from the initial frame. Meanwhile, Siamese trackers do not update network parameters online for real-time efficiency. The fixed target template and CNN parameters make Siamese trackers not effective to capture target appearance variations. In this paper, we propose a template updating method via reinforcement learning for Siamese regression trackers. We collect a series of templates and learn to maintain them based on an actor-critic framework. Among this framework, the actor network that is trained by deep reinforcement learning effectively updates the templates based on the tracking result on each frame. Besides the target template, we update the Siamese regression tracker online to adapt to target appearance variations. The experimental results on the standard benchmarks show the effectiveness of both template and network updating. The proposed tracker SiamRTU performs favorably against state-of-the-art approaches.

In this paper, we present an end-to-end learning framework for detailed 3D face reconstruction from a single image1. Our approach uses a 3DMM-based coarse model and a displacement map in UV-space to represent a 3D face. Unlike previous work addressing the problem, our learning framework does not require supervision of surrogate ground-truth 3D models computed with traditional approaches. Instead, we utilize the input image itself as supervision during learning. In the first stage, we combine a photometric loss and a facial perceptual loss between the input face and the rendered face, to regress a 3DMM-based coarse model. In the second stage, both the input image and the regressed texture of the coarse model are unwrapped into UV-space, and then sent through an image-toimage translation network to predict a displacement map in UVspace. The displacement map and the coarse model are used to render a final detailed face, which again can be compared with the original input image to serve as a photometric loss for the second stage. The advantage of learning displacement map in UV-space is that face alignment can be explicitly done during the unwrapping, thus facial details are easier to learn from large amount of data. Extensive experiments demonstrate the superiority of the proposed method over previous work.

Correlation filters (CF) have received considerable attention in visual tracking because of their computational efficiency. Leveraging deep features via off-the-shelf CNN models (e.g., VGG), CF trackers achieve state-of-the-art performance while consuming a large number of computing resources. This limits deep CF trackers to be deployed to many mobile platforms on which only a single-core CPU is available. In this paper, we propose to jointly compress and transfer off-the-shelf CNN models within a knowledge distillation framework. We formulate a CNN model pretrained from the image classification task as a teacher network, and distill this teacher network into a lightweight student network as the feature extractor to speed up CF trackers. In the distillation process, we propose a fidelity loss to enable the student network to maintain the representation capability of the teacher network. Meanwhile, we design a tracking loss to adapt the objective of the student network from object recognition to visual tracking. The distillation process is performed offline on multiple layers and adaptively updates the student network using a background-aware online learning scheme. The online adaptation stage exploits the background contents to improve the feature discrimination of the student network. Extensive experiments on six standard datasets demonstrate that the lightweight student network accelerates the speed of state-of-the-art deep CF trackers to real-time on a single-core CPU while maintaining almost the same tracking accuracy.

We propose an unsupervised visual tracking method in this paper. Different from existing approaches using extensive annotated data for supervised learning, our CNN model is trained on large-scale unlabeled videos in an unsupervised manner. Our motivation is that a robust tracker should be effective in both the forward and backward predictions (i.e., the tracker can forward localize the target object in successive frames and backtrace to its initial position in the first frame). We build our framework on a Siamese correlation filter network, which is trained using unlabeled raw videos. Meanwhile, we propose a multiple-frame validation method and a cost-sensitive loss to facilitate unsupervised learning. Without bells and whistles, the proposed unsupervised tracker achieves the baseline accuracy of fully supervised trackers, which require complete and accurate labels during training. Furthermore, unsupervised framework exhibits a potential in leveraging unlabeled or weakly labeled data to further improve the tracking accuracy.

We address the problem of recovering the 3D geometry of a human face from a set of facial images in multiple views. While recent studies have shown impressive progress in 3D Morphable Model (3DMM) based facial reconstruction, the settings are mostly restricted to a single view. There is an inherent drawback in the single-view setting: the lack of reliable 3D constraints can cause unresolvable ambiguities. We in this paper explore 3DMM-based shape recovery in a different setting, where a set of multi-view facial images are given as input. A novel approach is proposed to regress 3DMM parameters from multi-view inputs with an end-to-end trainable Convolutional Neural Network (CNN). Multi-view geometric constraints are incorporated into the network by establishing dense correspondences between different views leveraging a novel self-supervised view alignment loss. The main ingredient of the view alignment loss is a differentiable dense optical flow estimator that can backpropagate the alignment errors between an input view and a synthetic rendering from another input view, which is projected to the target view through the 3D shape to be inferred. Through minimizing the view alignment loss, better 3D shapes can be recovered such that the synthetic projections from one view to another can better align with the observed image. Extensive experiments demonstrate the superiority of the proposed method over other 3DMM methods.

The tracking-by-detection framework receives growing attention through the integration with the convolutional neural networks (CNNs). Existing tracking-by-detection-based methods, however, fail to track objects with severe appearance variations. This is because the traditional convolutional operation is performed on fixed grids, and thus may not be able to find the correct response while the object is changing pose or under varying environmental conditions. In this paper, we propose a deformable convolution layer to enrich the target appearance representations in the tracking-by-detection framework. We aim to capture the target appearance variations via deformable convolution, which adaptively enhances its original features. In addition, we also propose a gated fusion scheme to control how the variations captured by the deformable convolution affect the original appearance. The enriched feature representation through deformable convolution facilitates the discrimination of the CNN classifier on the target object and background. The extensive experiments on the standard benchmarks show that the proposed tracker performs favorably against the state-of-the-art methods.

The tracking-by-detection framework consists of two stages, i.e., drawing samples around the target object in the first stage and classifying each sample as the target object or as background in the second stage. The performance of existing trackers using deep classification networks is limited by two aspects. First, the positive samples in each frame are highly spatially overlapped, and they fail to capture rich appearance variations. Second, there exists extreme class imbalance between positive and negative samples. This paper presents the VITAL algorithm to address these two problems via adversarial learning. To augment positive samples, we use a generative network to randomly generate masks, which are applied to adaptively dropout input features to capture a variety of appearance changes. With the use of adversarial learning, our network identifies the mask that maintains the most robust features of the target objects over a long temporal span. In addition, to handle the issue of class imbalance, we propose a high-order cost sensitive loss to decrease the effect of easy negative samples to facilitate training the classification network. Extensive experiments on benchmark datasets demonstrate that the proposed tracker performs favorably against state-of-the-art approaches.

Image correction aims to adjust an input image into a visually pleasing one. Existing approaches are proposed mainly from the perspective of image pixel manipulation. They are not effective to recover the details in the under/over exposed regions. In this paper, we revisit the image formation procedure and notice that the missing details in these regions exist in the corresponding high dynamic range (HDR) data. These details are well perceived by the human eyes but diminished in the low dynamic range (LDR) domain because of the tone mapping process. Therefore, we formulate the image correction task as an HDR transformation process and propose a novel approach called Deep Reciprocating HDR Transformation (DRHT). Given an input LDR image, we first reconstruct the missing details in the HDR domain. We then perform tone mapping on the predicted HDR data to generate the output LDR image with the recovered details. To this end, we propose a united framework consisting of two CNNs for HDR reconstruction and tone mapping. They are integrated end-to-end for joint training and prediction. Experiments on the standard benchmarks demonstrate that the proposed method performs favorably against state-of-the-art image correction methods.

Due to the spatially variant blur caused by camera shake and object motions under different scene depths, deblurring images captured from dynamic scenes is challenging. Although recent works based on deep neural networks have shown great progress on this problem, their models are usually large and computationally expensive. In this paper, we propose a novel spatially variant neural network to address the problem. The proposed network is composed of three deep convolutional neural networks (CNNs) and a recurrent neural network (RNN). RNN is used as a deconvolution operator performed on feature maps extracted from the input image by one of the CNNs. Another CNN is used to learn the weights for the RNN at every location. As a result, the RNN is spatially variant and could implicitly model the deblurring process with spatially variant kernels. The third CNN is used to reconstruct the final deblurred feature maps into restored image. The whole network is end-to-end trainable. Our analysis shows that the proposed network has a large receptive field even with a small model size. Quantitative and qualitative evaluations on public datasets demonstrate that the proposed method performs favorably against state-of-the-art algorithms in terms of accuracy, speed, and model size.

Visual attention, derived from cognitive neuroscience, facilitates human perception on the most pertinent subset of the sensory data. Recently, significant efforts have been made to exploit attention schemes to advance computer vision systems. For visual tracking, it is often challenging to track target objects undergoing large appearance changes. Attention maps facilitate visual tracking by selectively paying attention to temporal robust features. Existing tracking-by-detection approaches mainly use additional attention modules to generate feature weights as the classifiers are not equipped with such mechanisms. In this paper, we propose a reciprocative learning algorithm to exploit visual attention for training deep classifiers. The proposed algorithm consists of feed-forward and backward operations to generate attention maps, which serve as regularization terms coupled with the original classification loss function for training. The deep classifier learns to attend to the regions of target objects robust to appearance changes. Extensive experiments on large-scale benchmark datasets show that the proposed attentive tracking method performs favorably against the state-of-the-art approaches.

Discriminative correlation filters (DCFs) have been shown to perform superiorly in visual tracking. They only need a small set of training samples from the initial frame to generate an appearance model. However, existing DCFs learn the filters separately from feature extraction, and update these filters using a moving average operation with an empirical weight. These DCF trackers hardly benefit from the end-to-end training. In this paper, we propose the CREST algorithm to reformulate DCFs as a one-layer convolutional neural network. Our method integrates feature extraction, response map generation as well as model update into the neural networks for an end-to-end training. To reduce model degradation during online update, we apply residual learning to take appearance changes into account. Extensive experiments on the benchmark datasets demonstrate that our CREST tracker performs favorably against state-of-the-art trackers.

Exemplar-based face sketch synthesis methods usually meet the challenging problem that input photos are captured in different lighting conditions from training photos. The critical step causing the failure is the search of similar patch candidates for an input photo patch. Conventional illumination invariant patch distances are adopted rather than directly relying on pixel intensity difference, but they will fail when local contrast within a patch changes. In this paper, we propose a fast preprocessing method named Bidirectional Luminance Remapping (BLR), which interactively adjust the lighting of training and input photos. Our method can be directly integrated into state-of-the-art exemplar-based methods to improve their robustness with ignorable computational cost

We propose a two-stage method for face hallucination. First, we generate facial components of the input image using CNNs. These components represent the basic facial structures. Second, we synthesize fine-grained facial structures from high resolution training images. The details of these structures are transferred into facial components for enhancement. Therefore, we generate facial components to approximate ground truth global appearance in the first stage and enhance them through recovering details in the second stage. The experiments demonstrate that our method performs favorably against state-of-the-art methods.

This paper presents a real-time decolorization method. Given the human visual systems preference for luminance information, the luminance should be preserved as much as possible during decolorization. As a result, the proposed decolorization method measures the amount of color contrast/detail lost when converting color to luminance. The detail loss is estimated by computing the difference between two intermediate images: one obtained by applying bilateral filter to the original color image, and the other obtained by applying joint bilateral filter to the original color image with its luminance as the guidance image. The estimated detail loss is then mapped to a grayscale image named residual image by minimizing the difference between the image gradients of the input color image and the objective grayscale image that is the sum of the residual image and the luminance. Apparently, the residual image will contain pixels with all zero values (that is the two intermediate images will be the same) only when no visual detail is missing in the luminance. Unlike most previous methods, the proposed decolorization method preserves both contrast in the color image and the luminance. Quantitative evaluation shows that it is the top performer on the standard test suite. Meanwhile it is very robust and can be directly used to convert videos while maintaining the temporal coherence. Specifically it can convert a high-resolution video (1280 × 720) in real time (about 28 Hz) on a 3.4 GHz i7 CPU.

We present a new efficient edge-preserving filter-"tree filter"-to achieve strong image smoothing. The proposed filter can smooth out high-contrast details while preserving major edges, which is not achievable for bilateral-filter-like techniques. Tree filter is a weighted-average filter, whose kernel is derived by viewing pixel affinity in a probabilistic framework simultaneously considering pixel spatial distance, color/intensity difference, as well as connectedness. Pixel connectedness is acquired by treating pixels as nodes in a minimum spanning tree (MST) extracted from the image. The fact that an MST makes all image pixels connected through the tree endues the filter with the power to smooth out high-contrast, fine-scale details while preserving major image structures, since pixels in small isolated region will be closely connected to surrounding majority pixels through the tree, while pixels inside large homogeneous region will be automatically dragged away from pixels outside the region. The tree filter can be separated into two other filters, both of which turn out to have fast algorithms. We also propose an efficient linear time MST extraction algorithm to further improve the whole filtering speed. The algorithms give tree filter a great advantage in low computational complexity (linear to number of image pixels) and fast speed: it can process a 1-megapixel 8-bit image at ~ 0.25 s on an Intel 3.4 GHz Core i7 CPU (including the construction of MST). The proposed tree filter is demonstrated on a variety of applications.

The state-of-the-art image retrieval approaches represent images with a high dimensional vector of visual words by quantizing local features, such as SIFT, in the descriptor space. The geometric clues among visual words in an image is usually ignored or exploited for full geometric verification, which is computationally expensive. In this paper, we focus on partial-duplicate web image retrieval, and propose a novel scheme, spatial coding, to encode the spatial relationships among local features in an image. Our spatial coding is both efficient and effective to discover false matches of local features between images, and can greatly improve retrieval performance. Experiments in partial-duplicate web image search, using a database of one million images, reveal that our approach achieves a 53% improvement in mean average precision and 46% reduction in time cost over the baseline bag-of-words approach.

The state-of-the-art image retrieval approaches represent images with a high dimensional vector of visual words by quantizing local features, such as SIFT, in the descriptor space. The geometric clues among visual words in an image is usually ignored or exploited for full geometric verification, which is computationally expensive. In recent years, partially duplicated images are prevalent on the web. In this demo, we focus on partial-duplicated web image retrieval, and propose a retrieval system based on a novel scheme, spatial coding, to encode the spatial information among local features in an image. Our spatial coding is both efficient and effective to discover false matches of local features between images, and can greatly improve retrieval performance.