Zeyu Wang

Zeyu Wang, Zhenzhen Weng, Serena Yeung-Levy

Continuous perception, the ability to integrate visual observations over time in a continuous stream fashion, is essential for robust real-world understanding, yet remains largely untested in current multimodal models. We introduce CP-Bench, a minimal and fully controlled benchmark designed to isolate this capability using an extremely simple task: counting identical cubes in a synthetic scene while the camera moves and only reveals subsets of objects at any moment. Despite the simplicity of the setting, we find that state-of-the-art open-source and commercial models, including Qwen-3-VL, InternVL3, GPT-5, and Gemini-3-Pro, fail dramatically. A static-camera control variant confirms that the failure arises not from object recognition but from an inability to accumulate evidence across time. Further experiments show that neither higher sampling FPS, perception- or spatial-enhanced models, nor finetuning with additional videos leads to meaningful cross-temporal generalization. Our results reveal a fundamental limitation in modern multimodal architectures and training paradigms. CP-Bench provides a simple yet powerful diagnostic tool and establishes a clean testbed for developing models capable of genuine time-consistent visual reasoning.

Zeyu Wang, Yu Wu

Retrieving target information based on input query is of fundamental importance in many real-world applications. In practice, it is not uncommon for the initial search to fail, where additional feedback information is needed to guide the searching process. In this work, we study a setting where the feedback is provided through users clicking liked and disliked searching results. We believe this form of feedback is of great practical interests for its convenience and efficiency. To facilitate future work in this direction, we construct a new benchmark termed click-feedback retrieval based on a large-scale dataset in fashion domain. We demonstrate that incorporating click-feedback can drastically improve the retrieval performance, which validates the value of the proposed setting. We also introduce several methods to utilize click-feedback during training, and show that click-feedback-guided training can significantly enhance the retrieval quality. We hope further exploration in this direction can bring new insights on building more efficient and user-friendly search engines.

Zeyu Wang, Haibin Shen, Changyou Men et al.

Image inpainting has achieved fundamental advances with deep learning. However, almost all existing inpainting methods aim to process natural images, while few target Thermal Infrared (TIR) images, which have widespread applications. When applied to TIR images, conventional inpainting methods usually generate distorted or blurry content. In this paper, we propose a novel task -- Thermal Infrared Image Inpainting, which aims to reconstruct missing regions of TIR images. Crucially, we propose a novel deep-learning-based model TIR-Fill. We adopt the edge generator to complete the canny edges of broken TIR images. The completed edges are projected to the normalization weights and biases to enhance edge awareness of the model. In addition, a refinement network based on gated convolution is employed to improve TIR image consistency. The experiments demonstrate that our method outperforms state-of-the-art image inpainting approaches on FLIR thermal dataset.

Kanghao Chen, Zeyu Wang, Lin Wang

Recent years have witnessed tremendous progress in the 3D reconstruction of dynamic humans from a monocular video with the advent of neural rendering techniques. This task has a wide range of applications, including the creation of virtual characters for virtual reality (VR) environments. However, it is still challenging to reconstruct clear humans when the monocular video is affected by motion blur, particularly caused by rapid human motion (e.g., running, dancing), as often occurs in the wild. This leads to distinct inconsistency of shape and appearance for the rendered 3D humans, especially in the blurry regions with rapid motion, e.g., hands and legs. In this paper, we propose ExFMan, the first neural rendering framework that unveils the possibility of rendering high-quality humans in rapid motion with a hybrid frame-based RGB and bio-inspired event camera. The ``out-of-the-box'' insight is to leverage the high temporal information of event data in a complementary manner and adaptively reweight the effect of losses for both RGB frames and events in the local regions, according to the velocity of the rendered human. This significantly mitigates the inconsistency associated with motion blur in the RGB frames. Specifically, we first formulate a velocity field of the 3D body in the canonical space and render it to image space to identify the body parts with motion blur. We then propose two novel losses, i.e., velocity-aware photometric loss and velocity-relative event loss, to optimize the neural human for both modalities under the guidance of the estimated velocity. In addition, we incorporate novel pose regularization and alpha losses to facilitate continuous pose and clear boundary. Extensive experiments on synthetic and real-world datasets demonstrate that ExFMan can reconstruct sharper and higher quality humans.

Zeyu Wang, Zhenzhen Weng, Serena Yeung-Levy

Conventional approaches to human mesh recovery predominantly employ a region-based strategy. This involves initially cropping out a human-centered region as a preprocessing step, with subsequent modeling focused on this zoomed-in image. While effective for single figures, this pipeline poses challenges when dealing with images featuring multiple individuals, as different people are processed separately, often leading to inaccuracies in relative positioning. Despite the advantages of adopting a whole-image-based approach to address this limitation, early efforts in this direction have fallen short in performance compared to recent region-based methods. In this work, we advocate for this under-explored area of modeling all people at once, emphasizing its potential for improved accuracy in multi-person scenarios through considering all individuals simultaneously and leveraging the overall context and interactions. We introduce a new model with a streamlined transformer-based design, featuring three critical design choices: multi-scale feature incorporation, focused attention mechanisms, and relative joint supervision. Our proposed model demonstrates a significant performance improvement, surpassing state-of-the-art region-based and whole-image-based methods on various benchmarks involving multiple individuals.

Xun Wei, Pukai Zhou, Zeyu Wang

The 2022 World Mental Health Report calls for global mental health care reform, amid rising prevalence of issues like anxiety and depression that affect nearly one billion people worldwide. Traditional in-person therapy fails to meet this demand, and the situation is worsened by stigma. While general-purpose large language models (LLMs) offer efficiency for AI-driven mental health solutions, they underperform because they lack specialized fine-tuning. Existing LLM-based mental health chatbots can engage in empathetic conversations, but they overlook real-time user mental state assessment which is critical for professional counseling. This paper proposes MoPHES, a framework that integrates mental state evaluation, conversational support, and professional treatment recommendations. The agent developed under this framework uses two fine-tuned MiniCPM4-0.5B LLMs: one is fine-tuned on mental health conditions datasets to assess users' mental states and predict the severity of anxiety and depression; the other is fine-tuned on multi-turn dialogues to handle conversations with users. By leveraging insights into users' mental states, our agent provides more tailored support and professional treatment recommendations. Both models are also deployed directly on mobile devices to enhance user convenience and protect user privacy. Additionally, to evaluate the performance of MoPHES with other LLMs, we develop a benchmark for the automatic evaluation of mental state prediction and multi-turn counseling dialogues, which includes comprehensive evaluation metrics, datasets, and methods.

Maoyu Wang, Yao Lu, Jiaqi Nie et al.

With the rapid development of deep learning, a growing number of pre-trained models have been publicly available. However, deploying these fixed models in real-world IoT applications is challenging because different devices possess heterogeneous computational and memory resources, making it impossible to deploy a single model across all platforms. Although traditional compression methods, such as pruning, quantization, and knowledge distillation, can improve efficiency, they become inflexible once applied and cannot adapt to changing resource constraints. To address these issues, we propose ReStNet, a Reusable and Stitchable Network that dynamically constructs a hybrid network by stitching two pre-trained models together. Implementing ReStNet requires addressing several key challenges, including how to select the optimal stitching points, determine the stitching order of the two pre-trained models, and choose an effective fine-tuning strategy. To systematically address these challenges and adapt to varying resource constraints, ReStNet determines the stitching point by calculating layer-wise similarity via Centered Kernel Alignment (CKA). It then constructs the hybrid model by retaining early layers from a larger-capacity model and appending deeper layers from a smaller one. To facilitate efficient deployment, only the stitching layer is fine-tuned. This design enables rapid adaptation to changing budgets while fully leveraging available resources. Moreover, ReStNet supports both homogeneous (CNN-CNN, Transformer-Transformer) and heterogeneous (CNN-Transformer) stitching, allowing to combine different model families flexibly. Extensive experiments on multiple benchmarks demonstrate that ReStNet achieve flexible accuracy-efficiency trade-offs at runtime while significantly reducing training cost.

Juncan Deng, Shuaiting Li, Zeyu Wang et al.

The rapid growth of the big neural network models puts forward new requirements for lightweight network representation methods. The traditional methods based on model compression have achieved great success, especially VQ technology which realizes the high compression ratio of models by sharing code words. However, because each layer of the network needs to build a code table, the traditional top-down compression technology lacks attention to the underlying commonalities, resulting in limited compression rate and frequent memory access. In this paper, we propose a bottom-up method to share the universal codebook among multiple neural networks, which not only effectively reduces the number of codebooks but also further reduces the memory access and chip area by storing static code tables in the built-in ROM. Specifically, we introduce VQ4ALL, a VQ-based method that utilizes codewords to enable the construction of various neural networks and achieve efficient representations. The core idea of our method is to adopt a kernel density estimation approach to extract a universal codebook and then progressively construct different low-bit networks by updating differentiable assignments. Experimental results demonstrate that VQ4ALL achieves compression rates exceeding 16 $\times$ while preserving high accuracy across multiple network architectures, highlighting its effectiveness and versatility.

Jaewoo Heo, George Hu, Zeyu Wang et al.

Human Mesh Recovery (HMR) is an important yet challenging problem with applications across various domains including motion capture, augmented reality, and biomechanics. Accurately predicting human pose parameters from a single image remains a challenging 3D computer vision task. In this work, we introduce DeforHMR, a novel regression-based monocular HMR framework designed to enhance the prediction of human pose parameters using deformable attention transformers. DeforHMR leverages a novel query-agnostic deformable cross-attention mechanism within the transformer decoder to effectively regress the visual features extracted from a frozen pretrained vision transformer (ViT) encoder. The proposed deformable cross-attention mechanism allows the model to attend to relevant spatial features more flexibly and in a data-dependent manner. Equipped with a transformer decoder capable of spatially-nuanced attention, DeforHMR achieves state-of-the-art performance for single-frame regression-based methods on the widely used 3D HMR benchmarks 3DPW and RICH. By pushing the boundary on the field of 3D human mesh recovery through deformable attention, we introduce an new, effective paradigm for decoding local spatial information from large pretrained vision encoders in computer vision.

Vidit Jain, Mukund Rungta, Yuchen Zhuang et al.

Hierarchical text classification (HTC) is a complex subtask under multi-label text classification, characterized by a hierarchical label taxonomy and data imbalance. The best-performing models aim to learn a static representation by combining document and hierarchical label information. However, the relevance of document sections can vary based on the hierarchy level, necessitating a dynamic document representation. To address this, we propose HiGen, a text-generation-based framework utilizing language models to encode dynamic text representations. We introduce a level-guided loss function to capture the relationship between text and label name semantics. Our approach incorporates a task-specific pretraining strategy, adapting the language model to in-domain knowledge and significantly enhancing performance for classes with limited examples. Furthermore, we present a new and valuable dataset called ENZYME, designed for HTC, which comprises articles from PubMed with the goal of predicting Enzyme Commission (EC) numbers. Through extensive experiments on the ENZYME dataset and the widely recognized WOS and NYT datasets, our methodology demonstrates superior performance, surpassing existing approaches while efficiently handling data and mitigating class imbalance. The data and code will be released publicly.

Zhenzhen Weng, Zeyu Wang, Serena Yeung

Recent advancements in text-to-image generation have enabled significant progress in zero-shot 3D shape generation. This is achieved by score distillation, a methodology that uses pre-trained text-to-image diffusion models to optimize the parameters of a 3D neural presentation, e.g. Neural Radiance Field (NeRF). While showing promising results, existing methods are often not able to preserve the geometry of complex shapes, such as human bodies. To address this challenge, we present ZeroAvatar, a method that introduces the explicit 3D human body prior to the optimization process. Specifically, we first estimate and refine the parameters of a parametric human body from a single image. Then during optimization, we use the posed parametric body as additional geometry constraint to regularize the diffusion model as well as the underlying density field. Lastly, we propose a UV-guided texture regularization term to further guide the completion of texture on invisible body parts. We show that ZeroAvatar significantly enhances the robustness and 3D consistency of optimization-based image-to-3D avatar generation, outperforming existing zero-shot image-to-3D methods.