Yanxiang Chen

Muyang Zheng, Yuanzhi Yao, Changting Lin et al.

Despite efforts to align large language models (LLMs) with societal and moral values, these models remain susceptible to jailbreak attacks -- methods designed to elicit harmful responses. Jailbreaking black-box LLMs is considered challenging due to the discrete nature of token inputs, restricted access to the target LLM, and limited query budget. To address the issues above, we propose an effective method for jailbreaking black-box large language Models via Iterative Semantic Tuning, named MIST. MIST enables attackers to iteratively refine prompts that preserve the original semantic intent while inducing harmful content. Specifically, to balance semantic similarity with computational efficiency, MIST incorporates two key strategies: sequential synonym search, and its advanced version -- order-determining optimization. We conduct extensive experiments on two datasets using two open-source and four closed-source models. Results show that MIST achieves competitive attack success rate, relatively low query count, and fair transferability, outperforming or matching state-of-the-art jailbreak methods. Additionally, we conduct analysis on computational efficiency to validate the practical viability of MIST.

Zhiyuan Wang, Yanxiang Chen, Yuanzhi Yao et al.

Detecting AI-generated images across unseen architectures remains challenging, as existing models often overfit to generator-specific fingerprints and semantic content rather than learning universal forgery traces. We attribute this failure to feature entanglement: detectors learn these factors as a single entangled representation, where universal forgery traces are inextricably confounded with both generator-specific fingerprints and semantic content. Crucially, our spectral analysis reveals that this entanglement is avoidable: distinct generator-specific fingerprints (e.g., GAN stripes vs. Diffusion Model spots) occupy disjoint frequency subspaces and coexist as independent superpositions. Leveraging this physical orthogonality, we propose the Orthogonal Decomposition and Purification Network (ODP-Net) to structurally disentangle these factors. Specifically, ODP-Net employs (1) Instance-aware Orthogonal Decomposition to project features into mutually exclusive subspaces: universal forgery traces, generator-specific fingerprints, and semantic content; (2) Perturbation-based Purification to enforce semantic invariance via cross-sample feature injection; and (3) Manifold Alignment to bridge domain gaps. By explicitly decoupling universal forgery traces from generator-specific fingerprints and semantic content, ODP-Net achieves state-of-the-art performance on unseen architectures (e.g., Stable Diffusion 3), validating that structural disentanglement is key to generalization.

Pengcheng Zhao, Jinxing Zhou, Yang Zhao et al.

The Audio-Visual Video Parsing task aims to recognize and temporally localize all events occurring in either the audio or visual stream, or both. Capturing accurate event semantics for each audio/visual segment is vital. Prior works directly utilize the extracted holistic audio and visual features for intra- and cross-modal temporal interactions. However, each segment may contain multiple events, resulting in semantically mixed holistic features that can lead to semantic interference during intra- or cross-modal interactions: the event semantics of one segment may incorporate semantics of unrelated events from other segments. To address this issue, our method begins with a Class-Aware Feature Decoupling (CAFD) module, which explicitly decouples the semantically mixed features into distinct class-wise features, including multiple event-specific features and a dedicated background feature. The decoupled class-wise features enable our model to selectively aggregate useful semantics for each segment from clearly matched classes contained in other segments, preventing semantic interference from irrelevant classes. Specifically, we further design a Fine-Grained Semantic Enhancement module for encoding intra- and cross-modal relations. It comprises a Segment-wise Event Co-occurrence Modeling (SECM) block and a Local-Global Semantic Fusion (LGSF) block. The SECM exploits inter-class dependencies of concurrent events within the same timestamp with the aid of a new event co-occurrence loss. The LGSF further enhances the event semantics of each segment by incorporating relevant semantics from more informative global video features. Extensive experiments validate the effectiveness of the proposed modules and loss functions, resulting in a new state-of-the-art parsing performance.

Zongmei Chen, Xin Liao, Xiaoshuai Wu et al.

The misuse of deepfake technology by malicious actors poses a potential threat to nations, societies, and individuals. However, existing methods for detecting deepfakes primarily focus on uncompressed videos, such as noise characteristics, local textures, or frequency statistics. When applied to compressed videos, these methods experience a decrease in detection performance and are less suitable for real-world scenarios. In this paper, we propose a deepfake video detection method based on 3D spatiotemporal trajectories. Specifically, we utilize a robust 3D model to construct spatiotemporal motion features, integrating feature details from both 2D and 3D frames to mitigate the influence of large head rotation angles or insufficient lighting within frames. Furthermore, we separate facial expressions from head movements and design a sequential analysis method based on phase space motion trajectories to explore the feature differences between genuine and fake faces in deepfake videos. We conduct extensive experiments to validate the performance of our proposed method on several compressed deepfake benchmarks. The robustness of the well-designed features is verified by calculating the consistent distribution of facial landmarks before and after video compression.Our method yields satisfactory results and showcases its potential for practical applications.

Pengcheng Zhao, Yanxiang Chen, Yang Zhao et al.

Automatic image colorization is inherently an ill-posed problem with uncertainty, which requires an accurate semantic understanding of scenes to estimate reasonable colors for grayscale images. Although recent interaction-based methods have achieved impressive performance, it is still a very difficult task to infer realistic and accurate colors for automatic colorization. To reduce the difficulty of semantic understanding of grayscale scenes, this paper tries to utilize corresponding audio, which naturally contains extra semantic information about the same scene. Specifically, a novel and pluggable audio-infused automatic image colorization (AIAIC) method is proposed, which consists of three stages. First, we take color image semantics as a bridge and pretrain a colorization network guided by color image semantics. Second, the natural co-occurrence of audio and video is utilized to learn the color semantic correlations between audio and visual scenes. Third, the implicit audio semantic representation is fed into the pretrained network to finally realize the audio-guided colorization. The whole process is trained in a self-supervised manner without human annotation. Experiments demonstrate that audio guidance can effectively improve the performance of automatic colorization, especially for some scenes that are difficult to understand only from visual modality.

Yanxiang Chen, Yuxing Hu, Luming Zhang et al.

Many aesthetic models in computer vision suffer from two shortcomings: 1) the low descriptiveness and interpretability of those hand-crafted aesthetic criteria (i.e., nonindicative of region-level aesthetics), and 2) the difficulty of engineering aesthetic features adaptively and automatically toward different image sets. To remedy these problems, we develop a deep architecture to learn aesthetically-relevant visual attributes from Flickr1, which are localized by multiple textual attributes in a weakly-supervised setting. More specifically, using a bag-ofwords (BoW) representation of the frequent Flickr image tags, a sparsity-constrained subspace algorithm discovers a compact set of textual attributes (e.g., landscape and sunset) for each image. Then, a weakly-supervised learning algorithm projects the textual attributes at image-level to the highly-responsive image patches at pixel-level. These patches indicate where humans look at appealing regions with respect to each textual attribute, which are employed to learn the visual attributes. Psychological and anatomical studies have shown that humans perceive visual concepts hierarchically. Hence, we normalize these patches and feed them into a five-layer convolutional neural network (CNN) to mimick the hierarchy of human perceiving the visual attributes. We apply the learned deep features on image retargeting, aesthetics ranking, and retrieval. Both subjective and objective experimental results thoroughly demonstrate the competitiveness of our approach.