Chunzhi Gu

Jianjian Qin, Chunzhi Gu, Jun Yu et al.

We present PD-REAL, a novel large-scale dataset for unsupervised anomaly detection (AD) in the 3D domain. It is motivated by the fact that 2D-only representations in the AD task may fail to capture the geometric structures of anomalies due to uncertainty in lighting conditions or shooting angles. PD-REAL consists entirely of Play-Doh models for 15 object categories and focuses on the analysis of potential benefits from 3D information in a controlled environment. Specifically, objects are first created with six types of anomalies, such as dent, crack, or perforation, and then photographed under different lighting conditions to mimic real-world inspection scenarios. To demonstrate the usefulness of 3D information, we use a commercially available RealSense camera to capture RGB and depth images. Compared to the existing 3D dataset for AD tasks, the data acquisition of PD-REAL is significantly cheaper, easily scalable and easier to control variables. Extensive evaluations with state-of-the-art AD algorithms on our dataset demonstrate the benefits as well as challenges of using 3D information. Our dataset can be downloaded from https://github.com/Andy-cs008/PD-REAL

Hiroto Nakata, Yawen Zou, Shunsuke Sakai et al.

Logical anomaly detection in industrial inspection remains challenging due to variations in visual appearance (e.g., background clutter, illumination shift, and blur), which often distract vision-centric detectors from identifying rule-level violations. However, existing benchmarks rarely provide controlled settings where logical states are fixed while such nuisance factors vary. To address this gap, we introduce VID-AD, a dataset for logical anomaly detection under vision-induced distraction. It comprises 10 manufacturing scenarios and five capture conditions, totaling 50 one-class tasks and 10,395 images. Each scenario is defined by two logical constraints selected from quantity, length, type, placement, and relation, with anomalies including both single-constraint and combined violations. We further propose a language-based anomaly detection framework that relies solely on text descriptions generated from normal images. Using contrastive learning with positive texts and contradiction-based negative texts synthesized from these descriptions, our method learns embeddings that capture logical attributes rather than low-level features. Extensive experiments demonstrate consistent improvements over baselines across the evaluated settings. The dataset is available at: https://github.com/nkthiroto/VID-AD.

Jianjian Qin, Chunzhi Gu, Jun Yu et al.

Anomaly detection, which is a critical and popular topic in computer vision, aims to detect anomalous samples that are different from the normal (i.e., non-anomalous) ones. The current mainstream methods focus on anomaly detection for images, whereas little attention has been paid to 3D point cloud. In this paper, drawing inspiration from the knowledge transfer ability of teacher-student architecture and the impressive feature extraction capability of recent neural networks, we design a teacher-student structured model for 3D anomaly detection. Specifically, we use feature space alignment, dimension zoom, and max pooling to extract the features of the point cloud and then minimize a multi-scale loss between the feature vectors produced by the teacher and the student networks. Moreover, our method only requires very few normal samples to train the student network due to the teacher-student distillation mechanism. Once trained, the teacher-student network pair can be leveraged jointly to fulfill 3D point cloud anomaly detection based on the calculated anomaly score. For evaluation, we compare our method against the reconstruction-based method on the ShapeNet-Part dataset. The experimental results and ablation studies quantitatively and qualitatively confirm that our model can achieve higher performance compared with the state of the arts in 3D anomaly detection with very few training samples.

Chunzhi Gu, Jun Yu, Chao Zhang

Generative model-based motion prediction techniques have recently realized predicting controlled human motions, such as predicting multiple upper human body motions with similar lower-body motions. However, to achieve this, the state-of-the-art methods require either subsequently learning mapping functions to seek similar motions or training the model repetitively to enable control over the desired portion of body. In this paper, we propose a novel framework to learn disentangled representations for controllable human motion prediction. Our network involves a conditional variational auto-encoder (CVAE) architecture to model full-body human motion, and an extra CVAE path to learn only the corresponding partial-body (e.g., lower-body) motion. Specifically, the inductive bias imposed by the extra CVAE path encourages two latent variables in two paths to respectively govern separate representations for each partial-body motion. With a single training, our model is able to provide two types of controls for the generated human motions: (i) strictly controlling one portion of human body and (ii) adaptively controlling the other portion, by sampling from a pair of latent spaces. Additionally, we extend and adapt a sampling strategy to our trained model to diversify the controllable predictions. Our framework also potentially allows new forms of control by flexibly customizing the input for the extra CVAE path. Extensive experimental results and ablation studies demonstrate that our approach is capable of predicting state-of-the-art controllable human motions both qualitatively and quantitatively.

Chunzhi Gu, Chao Zhang, Shigeru Kuriyama

The task of action-driven human motion prediction aims to forecast future human motion based on the observed sequence while respecting the given action label. It requires modeling not only the stochasticity within human motion but the smooth yet realistic transition between multiple action labels. However, the fact that most datasets do not contain such transition data complicates this task. Existing work tackles this issue by learning a smoothness prior to simply promote smooth transitions, yet doing so can result in unnatural transitions especially when the history and predicted motions differ significantly in orientations. In this paper, we argue that valid human motion transitions should incorporate realistic leg movements to handle orientation changes, and cast it as an action-conditioned in-betweening (ACB) learning task to encourage transition naturalness. Because modeling all possible transitions is virtually unreasonable, our ACB is only performed on very few selected action classes with active gait motions, such as Walk or Run. Specifically, we follow a two-stage forecasting strategy by first employing the motion diffusion model to generate the target motion with a specified future action, and then producing the in-betweening to smoothly connect the observation and prediction to eventually address motion prediction. Our method is completely free from the labeled motion transition data during training. To show the robustness of our approach, we generalize our trained in-betweening learning model on one dataset to two unseen large-scale motion datasets to produce natural transitions. Extensive experimental evaluations on three benchmark datasets demonstrate that our method yields the state-of-the-art performance in terms of visual quality, prediction accuracy, and action faithfulness.

Jianjian Qin, Chunzhi Gu, Jun Yu et al.

Anomaly detection (AD) is a fundamental task in computer vision. It aims to identify incorrect image data patterns which deviate from the normal ones. Conventional methods generally address AD by preparing augmented negative samples to enforce self-supervised learning. However, these techniques typically do not consider semantics during augmentation, leading to the generation of unrealistic or invalid negative samples. Consequently, the feature extraction network can be hindered from embedding critical features. In this study, inspired by visual attention learning approaches, we propose CutSwap, which leverages saliency guidance to incorporate semantic cues for augmentation. Specifically, we first employ LayerCAM to extract multilevel image features as saliency maps and then perform clustering to obtain multiple centroids. To fully exploit saliency guidance, on each map, we select a pixel pair from the cluster with the highest centroid saliency to form a patch pair. Such a patch pair includes highly similar context information with dense semantic correlations. The resulting negative sample is created by swapping the locations of the patch pair. Compared to prior augmentation methods, CutSwap generates more subtle yet realistic negative samples to facilitate quality feature learning. Extensive experimental and ablative evaluations demonstrate that our method achieves state-of-the-art AD performance on two mainstream AD benchmark datasets.

Wenqi Cai, Yawen Zou, Guang Li et al.

Dataset distillation (DD) aims to synthesize compact training sets that enable models to achieve high accuracy with significantly fewer samples. Recent diffusion-based DD methods commonly introduce semantic guidance through late-stage cross-attention, where textual prompts tend to dominate the generative process. Although this strategy enforces label relevance, it diminishes the contribution of visual latents, resulting in over-corrected samples that mirror prompt patterns rather than reflecting intrinsic visual features. To solve this problem, we introduce an Early Vision-Language Fusion (EVLF) method that aligns textual and visual embeddings at the transition between the encoder and the generative backbone. By incorporating a lightweight cross-attention module at this transition, the early representations simultaneously encode local textures and global semantic directions across the denoising process. Importantly, EVLF is plug-and-play and can be easily integrated into any diffusion-based dataset distillation pipeline with an encoder. It works across different denoiser architectures and sampling schedules without any task-specific modifications. Extensive experiments demonstrate that EVLF generates semantically faithful and visually coherent synthetic data, yielding consistent improvements in downstream classification accuracy across varied settings. Source code is available at https://github.com/wenqi-cai297/earlyfusion-for-dd/.

Shun Maeda, Chunzhi Gu, Koichiro Kamide et al.

Human-centric anomaly detection (AD) has been primarily studied to specify anomalous behaviors in a single person. However, as humans by nature tend to act in a collaborative manner, behavioral anomalies can also arise from human-human interactions. Detecting such anomalies using existing single-person AD models is prone to low accuracy, as these approaches are typically not designed to capture the complex and asymmetric dynamics of interactions. In this paper, we introduce a novel task, Human-Human Interaction Anomaly Detection (H2IAD), which aims to identify anomalous interactive behaviors within collaborative 3D human actions. To address H2IAD, we then propose Interaction Anomaly Detection Network (IADNet), which is formalized with a Temporal Attention Sharing Module (TASM). Specifically, in designing TASM, we share the encoded motion embeddings across both people such that collaborative motion correlations can be effectively synchronized. Moreover, we notice that in addition to temporal dynamics, human interactions are also characterized by spatial configurations between two people. We thus introduce a Distance-Based Relational Encoding Module (DREM) to better reflect social cues in H2IAD. The normalizing flow is eventually employed for anomaly scoring. Extensive experiments on human-human motion benchmarks demonstrate that IADNet outperforms existing Human-centric AD baselines in H2IAD.

Chunzhi Gu, Shigeru Kuriyama, Katsuya Hotta

Speech-driven facial animation aims to synthesize lip-synchronized 3D talking faces following the given speech signal. Prior methods to this task mostly focus on pursuing realism with deterministic systems, yet characterizing the potentially stochastic nature of facial motions has been to date rarely studied. While generative modeling approaches can easily handle the one-to-many mapping by repeatedly drawing samples, ensuring a diverse mode coverage of plausible facial motions on small-scale datasets remains challenging and less explored. In this paper, we propose predicting multiple samples conditioned on the same audio signal and then explicitly encouraging sample diversity to address diverse facial animation synthesis. Our core insight is to guide our model to explore the expressive facial latent space with a diversity-promoting loss such that the desired latent codes for diversification can be ideally identified. To this end, building upon the rich facial prior learned with vector-quantized variational auto-encoding mechanism, our model temporally queries multiple stochastic codes which can be flexibly decoded into a diverse yet plausible set of speech-faithful facial motions. To further allow for control over different facial parts during generation, the proposed model is designed to predict different facial portions of interest in a sequential manner, and compose them to eventually form full-face motions. Our paradigm realizes both diverse and controllable facial animation synthesis in a unified formulation. We experimentally demonstrate that our method yields state-of-the-art performance both quantitatively and qualitatively, especially regarding sample diversity.

Shun Maeda, Chunzhi Gu, Jun Yu et al.

We introduce the task of human action anomaly detection (HAAD), which aims to identify anomalous motions in an unsupervised manner given only the pre-determined normal category of training action samples. Compared to prior human-related anomaly detection tasks which primarily focus on unusual events from videos, HAAD involves the learning of specific action labels to recognize semantically anomalous human behaviors. To address this task, we propose a normalizing flow (NF)-based detection framework where the sample likelihood is effectively leveraged to indicate anomalies. As action anomalies often occur in some specific body parts, in addition to the full-body action feature learning, we incorporate extra encoding streams into our framework for a finer modeling of body subsets. Our framework is thus multi-level to jointly discover global and local motion anomalies. Furthermore, to show awareness of the potentially jittery data during recording, we resort to discrete cosine transformation by converting the action samples from the temporal to the frequency domain to mitigate the issue of data instability. Extensive experimental results on two human action datasets demonstrate that our method outperforms the baselines formed by adapting state-of-the-art human activity AD approaches to our task of HAAD.

Koichiro Kamide, Shunsuke Sakai, Shun Maeda et al.

Human Action Anomaly Detection (HAAD) aims to identify anomalous actions given only normal action data during training. Existing methods typically follow a one-model-per-category paradigm, requiring separate training for each action category and a large number of normal samples. These constraints hinder scalability and limit applicability in real-world scenarios, where data is often scarce or novel categories frequently appear. To address these limitations, we propose a unified framework for HAAD that is compatible with few-shot scenarios. Our method constructs a category-agnostic representation space via contrastive learning, enabling AD by comparing test samples with a given small set of normal examples (referred to as the support set). To improve inter-category generalization and intra-category robustness, we introduce a generative motion augmentation strategy harnessing a diffusion-based foundation model for creating diverse and realistic training samples. Notably, to the best of our knowledge, our work is the first to introduce such a strategy specifically tailored to enhance contrastive learning for action AD. Extensive experiments on the HumanAct12 dataset demonstrate the state-of-the-art effectiveness of our approach under both seen and unseen category settings, regarding training efficiency and model scalability for few-shot HAAD.

Yawen Zou, Guang Li, Zi Wang et al.

Dataset distillation (DD) aims to generate a compact yet informative dataset that achieves performance comparable to the original dataset, thereby reducing demands on storage and computational resources. Although diffusion models have made significant progress in dataset distillation, the generated surrogate datasets often contain samples with label inconsistencies or insufficient structural detail, leading to suboptimal downstream performance. To address these issues, we propose a detector-guided dataset distillation framework that explicitly leverages a pre-trained detector to identify and refine anomalous synthetic samples, thereby ensuring label consistency and improving image quality. Specifically, a detector model trained on the original dataset is employed to identify anomalous images exhibiting label mismatches or low classification confidence. For each defective image, multiple candidates are generated using a pre-trained diffusion model conditioned on the corresponding image prototype and label. The optimal candidate is then selected by jointly considering the detector's confidence score and dissimilarity to existing qualified synthetic samples, thereby ensuring both label accuracy and intra-class diversity. Experimental results demonstrate that our method can synthesize high-quality representative images with richer details, achieving state-of-the-art performance on the validation set.

Shunsuke Sakai, Xiangteng He, Chunzhi Gu et al.

Despite the remarkable success, recent reconstruction-based anomaly detection (AD) methods via diffusion modeling still involve fine-grained noise-strength tuning and computationally expensive multi-step denoising, leading to a fundamental tension between fidelity and efficiency. In this paper, we propose a novel inversion-based AD approach - detection via noising in latent space - which circumvents explicit reconstruction. Importantly, we contend that the limitations in prior reconstruction-based methods originate from the prevailing detection via denoising in RGB space paradigm. To address this, we model AD under a reconstruction-free formulation, which directly infers the final latent variable corresponding to the input image via DDIM inversion, and then measures the deviation based on the known prior distribution for anomaly scoring. Specifically, in approximating the original probability flow ODE using the Euler method, we only enforce very few inversion steps to noise the clean image to pursue inference efficiency. As the added noise is adaptively derived with the learned diffusion model, the original features for the clean testing image can still be leveraged to yield high detection accuracy. We perform extensive experiments and detailed analysis across three widely used image AD datasets under the unsupervised unified setting to demonstrate the effectiveness of our model, regarding state-of-the-art AD performance, and about 2 times inference time speedup without diffusion distillation.

Chunzhi Gu, Shuofeng Zhao, Chao Zhang

In this paper, we present a deep generative model based method to generate diverse human motion interpolation results. We resort to the Conditional Variational Auto-Encoder (CVAE) to learn human motion conditioned on a pair of given start and end motions, by leveraging the Recurrent Neural Network (RNN) structure for both the encoder and the decoder. Additionally, we introduce a regularization loss to further promote sample diversity. Once trained, our method is able to generate multiple plausible coherent motions by repetitively sampling from the learned latent space. Experiments on the publicly available dataset demonstrate the effectiveness of our method, in terms of sample plausibility and diversity.

Chunzhi Gu, Yan Zhao, Chao Zhang

Human motion prediction, which plays a key role in computer vision, generally requires a past motion sequence as input. However, in real applications, a complete and correct past motion sequence can be too expensive to achieve. In this paper, we propose a novel approach to predicting future human motions from a much weaker condition, i.e., a single image, with mixture density networks (MDN) modeling. Contrary to most existing deep human motion prediction approaches, the multimodal nature of MDN enables the generation of diverse future motion hypotheses, which well compensates for the strong stochastic ambiguity aggregated by the single input and human motion uncertainty. In designing the loss function, we further introduce the energy-based formulation to flexibly impose prior losses over the learnable parameters of MDN to maintain motion coherence as well as improve the prediction accuracy by customizing the energy functions. Our trained model directly takes an image as input and generates multiple plausible motions that satisfy the given condition. Extensive experiments on two standard benchmark datasets demonstrate the effectiveness of our method in terms of prediction diversity and accuracy.

Chunzhi Gu, Xuequan Lu, Chao Zhang

Color transfer, which plays a key role in image editing, has attracted noticeable attention recently. It has remained a challenge to date due to various issues such as time-consuming manual adjustments and prior segmentation issues. In this paper, we propose to model color transfer under a probability framework and cast it as a parameter estimation problem. In particular, we relate the transferred image with the example image under the Gaussian Mixture Model (GMM) and regard the transferred image color as the GMM centroids. We employ the Expectation-Maximization (EM) algorithm (E-step and M-step) for optimization. To better preserve gradient information, we introduce a Laplacian based regularization term to the objective function at the M-step which is solved by deriving a gradient descent algorithm. Given the input of a source image and an example image, our method is able to generate continuous color transfer results with increasing EM iterations. Various experiments show that our approach generally outperforms other competitive color transfer methods, both visually and quantitatively.

Chunzhi Gu, Xuequan Lu, Ying He et al.

Complex blur such as the mixup of space-variant and space-invariant blur, which is hard to model mathematically, widely exists in real images. In this paper, we propose a novel image deblurring method that does not need to estimate blur kernels. We utilize a pair of images that can be easily acquired in low-light situations: (1) a blurred image taken with low shutter speed and low ISO noise; and (2) a noisy image captured with high shutter speed and high ISO noise. Slicing the blurred image into patches, we extend the Gaussian mixture model (GMM) to model the underlying intensity distribution of each patch using the corresponding patches in the noisy image. We compute patch correspondences by analyzing the optical flow between the two images. The Expectation Maximization (EM) algorithm is utilized to estimate the parameters of GMM. To preserve sharp features, we add an additional bilateral term to the objective function in the M-step. We eventually add a detail layer to the deblurred image for refinement. Extensive experiments on both synthetic and real-world data demonstrate that our method outperforms state-of-the-art techniques, in terms of robustness, visual quality, and quantitative metrics.