arXiv subjects

Fei Zhao

Publications and source records attributed to Fei Zhao.

At least 37 records · Page 2Linked to original sources

arXiv2025

Research on Audio-Visual Quality Assessment Dataset and Method for User-Generated Omnidirectional Video

In response to the rising prominence of the Metaverse, omnidirectional videos (ODVs) have garnered notable interest, gradually shifting from professional-generated content (PGC) to user-generated content (UGC). However, the study of audio-visual quality assessment (AVQA) within ODVs remains limited. To address this, we construct a dataset of UGC omnidirectional audio and video (A/V) content. The videos are captured by five individuals using two different types of omnidirectional cameras, shooting 300 videos covering 10 different scene types. A subjective AVQA experiment is conducted on the dataset to obtain the Mean Opinion Scores (MOSs) of the A/V sequences. After that, to facilitate the development of UGC-ODV AVQA fields, we construct an effective AVQA baseline model on the proposed dataset, of which the baseline model consists of video feature extraction module, audio feature extraction and audio-visual fusion module. The experimental results demonstrate that our model achieves optimal performance on the proposed dataset.

cs.CV↗

arXiv2025

Towards Holistic Visual Quality Assessment of AI-Generated Videos: A LLM-Based Multi-Dimensional Evaluation Model

The development of AI-Generated Video (AIGV) technology has been remarkable in recent years, significantly transforming the paradigm of video content production. However, AIGVs still suffer from noticeable visual quality defects, such as noise, blurriness, frame jitter and low dynamic degree, which severely impact the user's viewing experience. Therefore, an effective automatic visual quality assessment is of great importance for AIGV content regulation and generative model improvement. In this work, we decompose the visual quality of AIGVs into three dimensions: technical quality, motion quality, and video semantics. For each dimension, we design corresponding encoder to achieve effective feature representation. Moreover, considering the outstanding performance of large language models (LLMs) in various vision and language tasks, we introduce a LLM as the quality regression module. To better enable the LLM to establish reasoning associations between multi-dimensional features and visual quality, we propose a specially designed multi-modal prompt engineering framework. Additionally, we incorporate LoRA fine-tuning technology during the training phase, allowing the LLM to better adapt to specific tasks. Our proposed method achieved \textbf{second place} in the NTIRE 2025 Quality Assessment of AI-Generated Content Challenge: Track 2 AI Generated video, demonstrating its effectiveness. Codes can be obtained at https://github.com/QiZelu/AIGVEval.

cs.CV↗

arXiv2025

NTIRE 2025 XGC Quality Assessment Challenge: Methods and Results

This paper reports on the NTIRE 2025 XGC Quality Assessment Challenge, which will be held in conjunction with the New Trends in Image Restoration and Enhancement Workshop (NTIRE) at CVPR 2025. This challenge is to address a major challenge in the field of video and talking head processing. The challenge is divided into three tracks, including user generated video, AI generated video and talking head. The user-generated video track uses the FineVD-GC, which contains 6,284 user generated videos. The user-generated video track has a total of 125 registered participants. A total of 242 submissions are received in the development phase, and 136 submissions are received in the test phase. Finally, 5 participating teams submitted their models and fact sheets. The AI generated video track uses the Q-Eval-Video, which contains 34,029 AI-Generated Videos (AIGVs) generated by 11 popular Text-to-Video (T2V) models. A total of 133 participants have registered in this track. A total of 396 submissions are received in the development phase, and 226 submissions are received in the test phase. Finally, 6 participating teams submitted their models and fact sheets. The talking head track uses the THQA-NTIRE, which contains 12,247 2D and 3D talking heads. A total of 89 participants have registered in this track. A total of 225 submissions are received in the development phase, and 118 submissions are received in the test phase. Finally, 8 participating teams submitted their models and fact sheets. Each participating team in every track has proposed a method that outperforms the baseline, which has contributed to the development of fields in three tracks.

cs.CV↗

arXiv2025

Room Impulse Response as a Prompt for Acoustic Echo Cancellation

Data-driven acoustic echo cancellation (AEC) methods, predominantly trained on synthetic or constrained real-world datasets, encounter performance declines in unseen echo scenarios, especially in real environments where echo paths are not directly observable. Our proposed method counters this limitation by integrating room impulse response (RIR) as a pivotal training prompt, aiming to improve the generalization of AEC models in such unforeseen conditions. We also explore four RIR prompt fusion methods. Comprehensive evaluations, including both simulated RIR under unknown conditions and recorded RIR in real, demonstrate that the proposed approach significantly improves performance compared to baseline models. These results substantiate the effectiveness of our RIR-guided approach in strengthening the model's generalization capabilities.

cs.SD↗

arXiv2025

Multi-Channel Acoustic Echo Cancellation Based on Direction-of-Arrival Estimation

Acoustic echo cancellation (AEC) is an important speech signal processing technology that can remove echoes from microphone signals to enable natural-sounding full-duplex speech communication. While single-channel AEC is widely adopted, multi-channel AEC can leverage spatial cues afforded by multiple microphones to achieve better performance. Existing multi-channel AEC approaches typically combine beamforming with deep neural networks (DNN). This work proposes a two-stage algorithm that enhances multi-channel AEC by incorporating sound source directional cues. Specifically, a lightweight DNN is first trained to predict the sound source directions, and then the predicted directional information, multi-channel microphone signals, and single-channel far-end signal are jointly fed into an AEC network to estimate the near-end signal. Evaluation results show that the proposed algorithm outperforms baseline approaches and exhibits robust generalization across diverse acoustic environments.

cs.SD↗

arXiv2025

CompBench: Benchmarking Complex Instruction-guided Image Editing

While real-world applications increasingly demand intricate scene manipulation, existing instruction-guided image editing benchmarks often oversimplify task complexity and lack comprehensive, fine-grained instructions. To bridge this gap, we introduce CompBench, a large-scale benchmark specifically designed for complex instruction-guided image editing. CompBench features challenging editing scenarios that incorporate fine-grained instruction following, spatial and contextual reasoning, thereby enabling comprehensive evaluation of image editing models' precise manipulation capabilities. To construct CompBench, we propose an MLLM-human collaborative framework with tailored task pipelines. Furthermore, we propose an instruction decoupling strategy that disentangles editing intents into four key dimensions: location, appearance, dynamics, and objects, ensuring closer alignment between instructions and complex editing requirements. Extensive evaluations reveal that CompBench exposes fundamental limitations of current image editing models and provides critical insights for the development of next-generation instruction-guided image editing systems. Our project page is available at https://comp-bench.github.io/.

cs.CV↗

arXiv2025

MarkMatch: Same-Hand Stuffing Detection

We present MarkMatch, a retrieval system for detecting whether two paper ballot marks were filled by the same hand. Unlike the previous SOTA method BubbleSig, which used binary classification on isolated mark pairs, MarkMatch ranks stylistic similarity between a query mark and a mark in the database using contrastive learning. Our model is trained with a dense batch similarity matrix and a dual loss objective. Each sample is contrasted against many negatives within each batch, enabling the model to learn subtle handwriting difference and improve generalization under handwriting variation and visual noise, while diagonal supervision reinforces high confidence on true matches. The model achieves an F1 score of 0.943, surpassing BubbleSig's best performance. MarkMatch also integrates Segment Anything Model for flexible mark extraction via box- or point-based prompts. The system offers election auditors a practical tool for visual, non-biometric investigation of suspicious ballots.

cs.CV↗

arXiv2025

Mitigating Image Captioning Hallucinations in Vision-Language Models

Hallucinations in vision-language models (VLMs) hinder reliability and real-world applicability, usually stemming from distribution shifts between pretraining data and test samples. Existing solutions, such as retraining or fine-tuning on additional data, demand significant computational resources and labor-intensive data collection, while ensemble-based methods incur additional costs by introducing auxiliary VLMs. To address these challenges, we propose a novel test-time adaptation framework using reinforcement learning to mitigate hallucinations during inference without retraining or any auxiliary VLMs. By updating only the learnable parameters in the layer normalization of the language model (approximately 0.003% of the model parameters), our method reduces distribution shifts between test samples and pretraining samples. A CLIP-based hallucination evaluation model is proposed to provide dual rewards to VLMs. Experimental results demonstrate a 15.4% and 17.3% reduction in hallucination rates on LLaVA and InstructBLIP, respectively. Our approach outperforms state-of-the-art baselines with a 68.3% improvement in hallucination mitigation, demonstrating its effectiveness.

cs.MM↗

arXiv2025

GenCLS++: Pushing the Boundaries of Generative Classification in LLMs Through Comprehensive SFT and RL Studies Across Diverse Datasets

As a fundamental task in machine learning, text classification plays a crucial role in many areas. With the rapid scaling of Large Language Models (LLMs), particularly through reinforcement learning (RL), there is a growing need for more capable discriminators. Consequently, advances in classification are becoming increasingly vital for enhancing the overall capabilities of LLMs. Traditional discriminative methods map text to labels but overlook LLMs' intrinsic generative strengths. Generative classification addresses this by prompting the model to directly output labels. However, existing studies still rely on simple SFT alone, seldom probing the interplay between training and inference prompts, and no work has systematically leveraged RL for generative text classifiers and unified SFT, RL, and inference-time prompting in one framework. We bridge this gap with GenCLS++, a framework that jointly optimizes SFT and RL while systematically exploring five high-level strategy dimensions-in-context learning variants, category definitions, explicit uncertainty labels, semantically irrelevant numeric labels, and perplexity-based decoding-during both training and inference. After an SFT "policy warm-up," we apply RL with a simple rule-based reward, yielding sizable extra gains. Across seven datasets, GenCLS++ achieves an average accuracy improvement of 3.46% relative to the naive SFT baseline; on public datasets, this improvement rises to 4.00%. Notably, unlike reasoning-intensive tasks that benefit from explicit thinking processes, we find that classification tasks perform better without such reasoning steps. These insights into the role of explicit reasoning provide valuable guidance for future LLM applications.

cs.CL↗

arXiv2025

Redefining Machine Translation on Social Network Services with Large Language Models

The globalization of social interactions has heightened the need for machine translation (MT) on Social Network Services (SNS), yet traditional models struggle with culturally nuanced content like memes, slang, and pop culture references. While large language models (LLMs) have advanced general-purpose translation, their performance on SNS-specific content remains limited due to insufficient specialized training data and evaluation benchmarks. This paper introduces RedTrans, a 72B LLM tailored for SNS translation, trained on a novel dataset developed through three innovations: (1) Supervised Finetuning with Dual-LLM Back-Translation Sampling, an unsupervised sampling method using LLM-based back-translation to select diverse data for large-scale finetuning; (2) Rewritten Preference Optimization (RePO), an algorithm that identifies and corrects erroneous preference pairs through expert annotation, building reliable preference corpora; and (3) RedTrans-Bench, the first benchmark for SNS translation, evaluating phenomena like humor localization, emoji semantics, and meme adaptation. Experiments show RedTrans outperforms state-of-the-art LLMs. Besides, RedTrans has already been deployed in a real-world production environment, demonstrating that domain-specific adaptation, effectively bridges the gap between generic and culturally grounded translation systems.

cs.CL↗

arXiv2025

Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models

DeepSeek-R1-Zero has successfully demonstrated the emergence of reasoning capabilities in LLMs purely through Reinforcement Learning (RL). Inspired by this breakthrough, we explore how RL can be utilized to enhance the reasoning capability of MLLMs. However, direct training with RL struggles to activate complex reasoning capabilities such as questioning and reflection in MLLMs, due to the absence of substantial high-quality multimodal reasoning data. To address this issue, we propose the reasoning MLLM, Vision-R1, to improve multimodal reasoning capability. Specifically, we first construct a high-quality multimodal CoT dataset without human annotations by leveraging an existing MLLM and DeepSeek-R1 through modality bridging and data filtering to obtain a 200K multimodal CoT dataset, Vision-R1-cold dataset. It serves as cold-start initialization data for Vision-R1. To mitigate the optimization challenges caused by overthinking after cold start, we propose Progressive Thinking Suppression Training (PTST) strategy and employ Group Relative Policy Optimization (GRPO) with the hard formatting result reward function to gradually refine the model's ability to learn correct and complex reasoning processes on a 10K multimodal math dataset. Comprehensive experiments show our model achieves an average improvement of $\sim$6% across various multimodal math reasoning benchmarks. Vision-R1-7B achieves a 73.5% accuracy on the widely used MathVista benchmark, which is only 0.4% lower than the leading reasoning model, OpenAI O1. Scaling up the amount of multimodal math data in the RL training, Vision-R1-32B and Vison-R1-72B achieves 76.4% and 78.2% MathVista benchmark scores, respectively. The datasets and code will be released in: https://github.com/Osilly/Vision-R1 .

cs.CV↗

arXiv2025

T2VEval: Benchmark Dataset and Objective Evaluation Method for T2V-generated Videos

Recent advances in text-to-video (T2V) technology, as demonstrated by models such as Runway Gen-3, Pika, Sora, and Kling, have significantly broadened the applicability and popularity of the technology. This progress has created a growing demand for accurate quality assessment metrics to evaluate the perceptual quality of T2V-generated videos and optimize video generation models. However, assessing the quality of text-to-video outputs remain challenging due to the presence of highly complex distortions, such as unnatural actions and phenomena that defy human cognition. To address these challenges, we constructed T2VEval-Bench, a multi-dimensional benchmark dataset for text-to-video quality evaluation, which contains 148 textual prompts and 1,783 videos generated by 13 T2V models. To ensure a comprehensive evaluation, we scored each video on four dimensions in the subjective experiment, which are overall impression, text-video consistency, realness, and technical quality. Based on T2VEval-Bench, we developed T2VEval, a multi-branch fusion scheme for T2V quality evaluation. T2VEval assesses videos across three branches: text-video consistency, realness, and technical quality. Using an attention-based fusion module, T2VEval effectively integrates features from each branch and predicts scores with the aid of a large language model. Additionally, we implemented a divide-and-conquer training strategy, enabling each branch to learn targeted knowledge while maintaining synergy with the others. Experimental results demonstrate that T2VEval achieves state-of-the-art performance across multiple metrics.

cs.CV↗

arXiv2024

GePBench: Evaluating Fundamental Geometric Perception for Multimodal Large Language Models

Geometric shapes play important roles in both physical world and human cognition. While multimodal large language models (MLLMs) have made significant advancements in visual understanding, their abilities to recognize geometric shapes and their spatial relationships, which we term \emph{geometric perception}, are not explicitly and systematically explored. To address this gap, we introduce GePBench, a novel benchmark specifically designed to assess the geometric perception capabilities of MLLMs. Our extensive evaluations reveal that even the current state-of-the-art MLLMs exhibit significant deficiencies in geometric perception tasks. Furthermore, we show that models trained with GePBench data demonstrate considerable improvements on a wide range of downstream tasks, highlighting the critical role of geometric perception in enabling advanced multimodal applications. Our code and datasets are available at \href{https://github.com/Changhao-Xiang/GePBench}{https://github.com/Changhao-Xiang/GePBench}.

cs.CL↗

arXiv2024

Attention-Enhanced Short-Time Wiener Solution for Acoustic Echo Cancellation

Acoustic Echo Cancellation (AEC) is an essential speech signal processing technology that removes echoes from microphone inputs to facilitate natural-sounding full-duplex communication. Currently, deep learning-based AEC methods primarily focus on refining model architectures, frequently neglecting the incorporation of knowledge from traditional filter theory. This paper presents an innovative approach to AEC by introducing an attention-enhanced short-time Wiener solution. Our method strategically harnesses attention mechanisms to mitigate the impact of double-talk interference, thereby optimizing the efficiency of knowledge utilization. The derivation of the short-term Wiener solution, which adapts classical Wiener solutions to finite input causality, integrates established insights from filter theory into this method. The experimental outcomes corroborate the effectiveness of our proposed approach, surpassing other baseline models in performance and generalization. The official code is available at https://github.com/ZhaoF-i/ASTWS-AEC

cs.SD↗

arXiv2024

Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification

Multimodal Large Language Models (MLLMs) have achieved remarkable success in vision understanding, reasoning, and interaction. However, the inference computation and memory increase progressively with the generation of output tokens during decoding, directly affecting the efficacy of MLLMs. Existing methods attempt to reduce the vision context redundancy to achieve efficient MLLMs. Unfortunately, the efficiency benefits of the vision context reduction in the prefill stage gradually diminish during the decoding stage. To address this problem, we proposed a dynamic vision-language context sparsification framework Dynamic-LLaVA, which dynamically reduces the redundancy of vision context in the prefill stage and decreases the memory and computation overhead of the generated language context during decoding. Dynamic-LLaVA designs a tailored sparsification inference scheme for different inference modes, i.e., prefill, decoding with and without KV cache, to achieve efficient inference of MLLMs. In practice, Dynamic-LLaVA can reduce computation consumption by $\sim$75\% in the prefill stage. Meanwhile, throughout the entire generation process of MLLMs, Dynamic-LLaVA reduces the $\sim$50\% computation consumption under decoding without KV cache, while saving $\sim$50\% GPU memory overhead when decoding with KV cache, due to the vision-language context sparsification. Extensive experiments also demonstrate that Dynamic-LLaVA achieves efficient inference for MLLMs with negligible understanding and generation ability degradation or even performance gains compared to the full-context inference baselines. Code is available at https://github.com/Osilly/dynamic_llava .

cs.CV↗

arXiv2024

CHESTNUT: A QoS Dataset for Mobile Edge Environments

Quality of Service (QoS) is an important metric to measure the performance of network services. Nowadays, it is widely used in mobile edge environments to evaluate the quality of service when mobile devices request services from edge servers. QoS usually involves multiple dimensions, such as bandwidth, latency, jitter, and data packet loss rate. However, most existing QoS datasets, such as the common WS-Dream dataset, focus mainly on static QoS metrics of network services and ignore dynamic attributes such as time and geographic location. This means they should have detailed the mobile device's location at the time of the service request or the chronological order in which the request was made. However, these dynamic attributes are crucial for understanding and predicting the actual performance of network services, as QoS performance typically fluctuates with time and geographic location. To this end, we propose a novel dataset that accurately records temporal and geographic location information on quality of service during the collection process, aiming to provide more accurate and reliable data to support future QoS prediction in mobile edge environments.

cs.LG↗

arXiv2024

AlignGPT: Multi-modal Large Language Models with Adaptive Alignment Capability

Multimodal Large Language Models (MLLMs) are widely regarded as crucial in the exploration of Artificial General Intelligence (AGI). The core of MLLMs lies in their capability to achieve cross-modal alignment. To attain this goal, current MLLMs typically follow a two-phase training paradigm: the pre-training phase and the instruction-tuning phase. Despite their success, there are shortcomings in the modeling of alignment capabilities within these models. Firstly, during the pre-training phase, the model usually assumes that all image-text pairs are uniformly aligned, but in fact the degree of alignment between different image-text pairs is inconsistent. Secondly, the instructions currently used for finetuning incorporate a variety of tasks and different tasks usually require different levels of alignment capabilities, but previous MLLMs overlook these differentiated alignment needs. To tackle these issues, we propose a new multimodal large language model AlignGPT. In the pre-training stage, instead of treating all image-text pairs equally, we divide them into different groups according to the degrees of alignment of them. Then, the model is trained to learn the representations of different alignment levels. In the instruction-tuning phase, we adaptively combine these representations of alignment levels to meet the dynamic alignment needs of different tasks. Extensive experimental results show that our model achieves competitive performance on 12 benchmarks.

cs.CL↗

arXiv2024

Knowledge-aware Dual-side Attribute-enhanced Recommendation

\textit{Knowledge-aware} recommendation methods (KGR) based on \textit{graph neural networks} (GNNs) and \textit{contrastive learning} (CL) have achieved promising performance. However, they fall short in modeling fine-grained user preferences and further fail to leverage the \textit{preference-attribute connection} to make predictions, leading to sub-optimal performance. To address the issue, we propose a method named \textit{\textbf{K}nowledge-aware \textbf{D}ual-side \textbf{A}ttribute-enhanced \textbf{R}ecommendation} (KDAR). Specifically, we build \textit{user preference representations} and \textit{attribute fusion representations} upon the attribute information in knowledge graphs, which are utilized to enhance \textit{collaborative filtering} (CF) based user and item representations, respectively. To discriminate the contribution of each attribute in these two types of attribute-based representations, a \textit{multi-level collaborative alignment contrasting} mechanism is proposed to align the importance of attributes with CF signals. Experimental results on four benchmark datasets demonstrate the superiority of KDAR over several state-of-the-art baselines. Further analyses verify the effectiveness of our method. The code of KDAR is released at: \href{https://github.com/TJTP/KDAR}{https://github.com/TJTP/KDAR}.

cs.IR↗