Author name cluster

Yuhao Chen

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

AAAI Conference 2026 Conference Paper

Look as You Think: Unifying Reasoning and Visual Evidence Attribution for Verifiable Document RAG via Reinforcement Learning

Shuochen Liu
Pengfei Luo
Chao Zhang
Yuhao Chen
Haotian Zhang
Qi Liu
Xin Kou
Tong Xu

Aiming to identify precise evidence sources from visual documents, visual evidence attribution for visual document retrieval–augmented generation (VD-RAG) ensures reliable and verifiable predictions from vision-language models (VLMs) in multimodal question answering. Most existing methods adopt end-to-end training to facilitate intuitive answer verification. However, they lack fine-grained supervision and progressive traceability throughout the reasoning process. In this paper, we introduce the Chain-of-Evidence (CoE) paradigm for VD-RAG. CoE unifies Chain-of-Thought (CoT) reasoning and visual evidence attribution by grounding reference elements in reasoning steps to specific regions with bounding boxes and page indexes. To enable VLMs to generate such evidence-grounded reasoning, we propose Look As You Think (LAT), a reinforcement learning framework that trains models to produce verifiable reasoning paths with consistent attribution. During training, LAT evaluates the attribution consistency of each evidence region and provides rewards only when the CoE trajectory yields correct answers, encouraging process-level self-verification. Experiments on vanilla Qwen2.5-VL-7B-Instruct with Paper‑ and Wiki‑VISA benchmarks show that LAT consistently improves the vanilla model in both single- and multi-image settings, yielding average gains of 8.23% in soft exact match (EM) and 47.0% in IoU@0.5. Meanwhile, LAT not only outperforms the supervised fine-tuning baseline, which is trained to directly produce answers with attribution, but also exhibits stronger generalization across domains.

PDF Details DOI

AAAI Conference 2026 Conference Paper

MosaicDoc: A Large-Scale Bilingual Benchmark for Visually Rich Document Understanding

Ketong Chen
Yuhao Chen
Yang Xue

Despite the rapid progress of Vision-Language Models (VLMs), their capabilities are inadequately assessed by existing benchmarks, which are predominantly English-centric, feature simplistic layouts, and support limited tasks. Consequently, they fail to evaluate model performance for Visually Rich Document Understanding (VRDU), a critical challenge involving complex layouts and dense text. To address this, we introduce DocWeaver, a novel multi-agent pipeline that leverages Large Language Models to automatically generate a new benchmark. The result is MosaicDoc, a large-scale, bilingual (Chinese and English) resource designed to push the boundaries of VRDU. Sourced from newspapers and magazines, MosaicDoc features diverse and complex layouts (including multi-column and non-Manhattan), rich stylistic variety from 196 publishers, and comprehensive multi-task annotations (OCR, VQA, reading order, and localization). With 72K images and over 600K QA pairs, MosaicDoc serves as a definitive benchmark for the field. Our extensive evaluation of state-of-the-art models on this benchmark reveals their current limitations in handling real-world document complexity and charts a clear path for future research.

PDF Details DOI

JBHI Journal 2026 Journal Article

Physically Informed 3D Food Reconstruction: Methods and Results

Jiangpeng He
Yuhao Chen
Gautham Vinod
Xiaoyan Zhang
Talha Ibn Mahmud
Ahmad AlMughrabi
Umair Haroon
Ricardo Marques

Accurate food portion size estimation is a critical challenge in nutrition analysis and dietary assessment. Recent 3D reconstruction methods primarily focus on surface geometry, often neglecting the volumetric accuracy necessary for precise portion size estimation. In this work, we present three methods, NS-DRS, HR-CMS, and GS-GP, for real-scale 3D food reconstruction and volume estimation from limited 2D inputs. All three methods follow a shared three-stage pipeline consisting of 3D reconstruction, scale estimation using physical references such as visible checkerboards, and mesh refinement. While this high-level structure is consistent, the methods differ in their reconstruction backbones and scale-recovery mechanisms, and are designed for different input conditions, including both single-view and multi-view images. We evaluate each method on a diverse set of food items exhibiting variations in texture, shape, and camera pose, and assess their performance in terms of volume estimation and geometric accuracy. The results reveal complementary strengths among the pipelines, with NS-DRS performing better in volume estimation and GS-GP being more effective in 3D reconstruction. Across all settings, the three methods achieve 18–23% lower volume estimation error compared to the current state-of-the-art model. These findings demonstrate the effectiveness of physically informed and explainable reconstruction pipelines for accurate portion estimation, and support their potential use as scalable tools for dietary monitoring and clinical nutrition analysis.

Details DOI

ICML Conference 2025 Conference Paper

Asymmetric Decision-Making in Online Knowledge Distillation: Unifying Consensus and Divergence

Zhaowei Chen
Borui Zhao
Yuchen Ge
Yuhao Chen
Renjie Song
Jiajun Liang

Online Knowledge Distillation (OKD) methods represent a streamlined, one-stage distillation training process that obviates the necessity of transferring knowledge from a pretrained teacher network to a more compact student network. In contrast to existing logits-based OKD methods, this paper presents an innovative approach to leverage intermediate spatial representations. Our analysis of the intermediate features from both teacher and student models reveals two pivotal insights: (1) the similar features between students and teachers are predominantly focused on the foreground objects. (2) teacher models emphasize foreground objects more than students. Building on these findings, we propose Asymmetric Decision-Making (ADM) to enhance feature consensus learning for student models while continuously promoting feature diversity in teacher models. Specifically, Consensus Learning for student models prioritizes spatial features with high consensus relative to teacher models. Conversely, Divergence Learning for teacher models highlights spatial features with lower similarity compared to student models, indicating superior performance by teacher models in these regions. Consequently, ADM facilitates the student models to catch up with the feature learning process of the teacher models. Extensive experiments demonstrate that ADM consistently surpasses existing OKD methods across various online knowledge distillation settings and also achieves superior results when transferred to offline knowledge distillation, semantic segmentation and diffusion distillation tasks.

Details

NeurIPS Conference 2025 Conference Paper

LEDiT: Your Length-Extrapolatable Diffusion Transformer without Positional Encoding

Shen Zhang
Siyuan Liang
Yaning Tan
Zhaowei Chen
Linze Li
Ge Wu
Yuhao Chen
Shuheng Li

Diffusion transformers (DiTs) struggle to generate images at resolutions higher than their training resolutions. The primary obstacle is that the explicit positional encodings (PE), such as RoPE, need extrapolating to unseen positions which degrades performance when the inference resolution differs from training. In this paper, We propose a Length-Extrapolatable Diffusion Transformer (LEDiT) to overcome this limitation. LEDiT needs no explicit PEs, thereby avoiding PE extrapolation. The key innovation of LEDiT lies in the use of causal attention. We demonstrate that causal attention can implicitly encode global positional information and show that such information facilitates extrapolation. We further introduce a locality enhancement module, which captures fine-grained local information to complement the global coarse-grained position information encoded by causal attention. Experimental results on both conditional and text-to-image generation tasks demonstrate that LEDiT supports up to 4× resolution scaling (e. g. , from 256$\times$256 to 512$\times$512), achieving better image quality compared to the state-of-the-art length extrapolation methods. We believe that LEDiT marks a departure from the standard RoPE-based methods and offers a promising insight into length extrapolation. Project page: https: //shenzhang2145. github. io/ledit/

PDF Details

NeurIPS Conference 2024 Conference Paper

Seeing Beyond the Crop: Using Language Priors for Out-of-Bounding Box Keypoint Prediction

Bavesh Balaji
Jerrin Bright
Yuhao Chen
Sirisha Rambhatla
John S. Zelek
David A. Clausi

Accurate estimation of human pose and the pose of interacting objects, like a hockey stick, is crucial for action recognition and performance analysis, particularly in sports. Existing methods capture the object along with the human in the bounding boxes, assuming all keypoints are visible within the bounding box. This necessitates larger bounding boxes to capture the object, introducing unnecessary visual features and hindering performance in real-world cluttered environments. We propose a simple image and text-based multimodal solution TokenCLIPose that addresses this limitation. Our approach focuses solely on human keypoints within the bounding box, treating objects as unseen. TokenCLIPose leverages the rich semantic representations endowed by language for inducing keypoint-specific context, even for occluded keypoints. We evaluate the performance of TokenCLIPose on a real-world Ice-Hockey dataset, and demonstrate its generalizability through zero-shot transfer to a smaller Lacrosse dataset. Additionally, we showcase its flexibility on CrowdPose, a popular occlusion benchmark with keypoints within the bounding box. Our method significantly improves over state-of-the-art approaches on all three datasets, with gains of 4. 36\%, 2. 35\%, and 3. 8\%, respectively.

PDF Details DOI

JBHI Journal 2023 Journal Article

sEMG-Based End-to-End Continues Prediction of Human Knee Joint Angles Using the Tightly Coupled Convolutional Transformer Model

Tuanjie Liang
Ning Sun
Qiong Wang
Jingyu Bu
Long Li
Yuhao Chen
Menglin Cao
Jin Ma

Wearable exoskeleton robots can promote the rehabilitation of patients with physical dysfunction. And improving human-computer interaction performance is a significant challenge for exoskeleton robots. The traditional feature extraction process based on surface Electromyography(sEMG) is complex and requires manual intervention, making real-time performance difficult to guarantee. In this study, we propose an end-to-end method to predict human knee joint angles based on sEMG signals using a tightly coupled convolutional transformer (TCCT) model. We first collected sEMG signals from 5 healthy subjects. Then, the envelope was extracted from the noise-removed sEMG signal and used as the input to the model. Finally, we developed the TCCT model to predict the knee joint angle after 100 ms. For the prediction performance, we used the Root Mean Square Error(RMSE), Pearson Correlation Coefficient(CC), and Adjustment R 2 as metrics to evaluate the error between the actual knee angle and the predicted knee angle. The results show that the model can predict the human knee angle quickly and accurately. The mean RMSE, Adjustment R 2, and (CC) values of the model are 3. 79°, 0. 96, and 0. 98, respectively, which are better than traditional deep learning models such as Informer (4. 14, 0. 95, 0. 98), CNN (5. 56, 0. 89, 0. 96) and CNN-BiLSTM (3. 97, 0. 95, 0. 98). In addition, the prediction time of our proposed model is only 11. 67 ± 0. 67 ms, which is less than 100 ms. Therefore, the real-time and accuracy of the model can meet the continuous prediction of human knee joint angle in practice.

Details DOI

IJCAI Conference 2021 Conference Paper

Low Resolution Information Also Matters: Learning Multi-Resolution Representations for Person Re-Identification

Guoqing Zhang
Yuhao Chen
Weisi Lin
Arun Chandran
Xuan Jing

As a prevailing task in video surveillance and forensics field, person re-identification (re-ID) aims to match person images captured from non-overlapped cameras. In unconstrained scenarios, person images often suffer from the resolution mismatch problem, i. e. , Cross-Resolution Person Re-ID. To overcome this problem, most existing methods restore low resolution (LR) images to high resolution (HR) by super-resolution (SR). However, they only focus on the HR feature extraction and ignore the valid information from original LR images. In this work, we explore the influence of resolutions on feature extraction and develop a novel method for cross-resolution person re-ID called Multi-Resolution Representations Joint Learning (MRJL). Our method consists of a Resolution Reconstruction Network (RRN) and a Dual Feature Fusion Network (DFFN). The RRN uses an input image to construct a HR version and a LR version with an encoder and two decoders, while the DFFN adopts a dual-branch structure to generate person representations from multi-resolution images. Comprehensive experiments on five benchmarks verify the superiority of the proposed MRJL over the relevent state-of-the-art methods.

PDF Details DOI