arXiv subjects

Zixuan Zou

Publications and source records attributed to Zixuan Zou.

4 recordsLinked to original sources

arXiv2026

Depth-Guided Video Object Counting in Crowded Scenes

Our primary objective is to advance video object counting in crowded scenes, aiming to robustly count all instances of a target category based on given text or visual prompts. Existing methods rely on RGB information, limiting their discriminative ability in crowded and occluded conditions. To address this, we propose a Depth-Guided Detector (DG-Det) along with a general post-processing pipeline. By integrating depth cues with multi-scale RGB-D cross-attention and explicit occlusion prediction, our method enhances spatial understanding and achieves robust detection in crowded and occluded scenes. Furthermore, we introduce a unified de-duplication framework to eliminate cross-frame redundant counting. To facilitate future research, we also release a new RGB-D Video Object Counting dataset featuring depth information and multiple object categories persequence. Extensive experiments demonstrate that our method achieves a 62.01\% reduction in MAE compared to existing baselines, and also produces consistent improvements in RMSE. We provide the source code at https://github.com/streamer-AP/DG-Net and the dataset at https://huggingface.co/datasets/aerospace123/RGBD-VideoCount.

cs.CV↗

arXiv2026

EmoStyle: Affective Conditioning of Style-Specialist Experts for Emotional Image Generation

Emotion-aware artistic image generation requires an image to match the input prompt, follow the specified artistic style, and convey the target emotion. In this challenge, the main difficulty is that the visual and affective attributes available in the training data are not explicitly provided at test time. Without these attributes, the generator has to decide not only what to depict, but also how the target emotion should be expressed through color, lighting, brushwork, composition, line, and layout. This creates a control gap between the available test prompt and the fine-grained conditions needed for emotion-aware artistic generation. To bridge this gap, we propose EmoStyle, a Z-Image-based framework that converts the input prompt into a structured generation state. An LLM reasoner first predicts affective cues (valence-arousal, dominant emotion, and therapeutic-effect labels) and an aspect-ratio decision. Instead of using these predictions only as additional prompt text, we encode the affective fields into an affective condition vector and inject it into the denoising blocks through AdaLN-style modulation. This allows the inferred control variables to directly guide the generation of intermediate features. Since emotional expression is also style-dependent, we further train a dedicated LoRA adapter for each artistic style bucket and select the corresponding expert during inference, enabling the same affective cues to be rendered with bucket-specific priors for color, texture, brushwork, and composition. Finally, a lightweight VLM-guided candidate selection step ranks the generated images based on prompt alignment, style consistency, emotional expression, and visual quality. In Track 1 of the AffectiveArt Challenge 2026, our USTC\_PI\_LAB\_TEAM submission achieved first place.

cs.CV↗

arXiv2025

High Chern numbers and topological flat bands in high-field polarized Kitaev magnets on the star lattice

The geometrically frustrated Kitaev magnets are demonstrated to be fertile playgrounds that allow for the occurrence of exotic phenomena, including topological phases and the thermal Hall effect. Notwithstanding the established consensus that the field-polarized phase in the honeycomb-lattice Kitaev magnet hosts topological magnons exhibiting Chern numbers $C = \pm1$, the nature of magnon excitations in Kitaev magnets on the star lattice, a triangle-decorated honeycomb lattice, has rarely been explored primarily due to its complicated geometry. To this end, we study the band topology of magnons on the star lattice in the presence of a strong out-of-plane magnetic field using linear spin-wave theory. By calculating the Chern numbers of magnon bands, we find that topological phase diagrams are predominantly composed of two distinct topological phases whose Chern numbers are different by a sign in inverse order. Remarkably, each phase is characterized by a high Chern number of either $+2$ or $-2$. In addition, several topological flat bands with large flatness are identified. The two phases are separated by a dozen narrow topological high-Chern-number segments, whose region shrinks as the magnetic field increases and vanishes eventually. We also find that the thermal Hall conductivity approaches zero at certain parameters, and it changes (keeps) its sign when crossing the topological phase-transition points (flat-band points).

cond-mat.str-el↗

arXiv2022

Aiming in Harsh Environments: A New Framework for Flexible and Adaptive Resource Management

The harsh environment imposes a unique set of challenges on networking strategies. In such circumstances, the environmental impact on network resources and long-time unattended maintenance has not been well investigated yet. To address these challenges, we propose a flexible and adaptive resource management framework that incorporates the environment awareness functionality. In particular, we propose a new network architecture and introduce the new functionalities against the traditional network components. The novelties of the proposed architecture include a deep-learning-based environment resource prediction module and a self-organized service management module. Specifically, the available network resource under various environmental conditions is predicted by using the prediction module. Then based on the prediction, an environment-oriented resource allocation method is developed to optimize the system utility. To demonstrate the effectiveness and efficiency of the proposed new functionalities, we examine the method via an experiment in a case study. Finally, we introduce several promising directions of resource management in harsh environments that can be extended from this paper.

cs.NI↗