Xiaohui Li

Shuo Cao, Nan Ma, Jiayang Li et al.

The rapid advancement of educational applications, artistic creation, and AI-generated content (AIGC) technologies has substantially increased practical requirements for comprehensive Image Aesthetics Assessment (IAA), particularly demanding methods capable of delivering both quantitative scoring and professional understanding. Multimodal Large Language Model (MLLM)-based IAA methods demonstrate stronger perceptual and generalization capabilities compared to traditional approaches, yet they suffer from modality bias (score-only or text-only) and lack fine-grained attribute decomposition, thereby failing to support further aesthetic assessment. In this paper, we present:(1) ArtiMuse, an innovative MLLM-based IAA model with Joint Scoring and Expert-Level Understanding capabilities; (2) ArtiMuse-10K, the first expert-curated image aesthetic dataset comprising 10,000 images spanning 5 main categories and 15 subcategories, each annotated by professional experts with 8-dimensional attributes analysis and a holistic score. Both the model and dataset will be made public to advance the field.

Shuai Li, Jian Xu, Xiao-Hui Li et al.

Recent advances in Multi-modal Large Language Models (MLLMs) have shown significant progress in open-world Visual Question Answering (VQA). However, integrating visual information increases the number of processed tokens, leading to higher GPU memory usage and computational overhead. Images often contain more redundant information than text, and not all visual details are pertinent to specific questions. To address these challenges, we propose QG-VTC, a novel question-guided visual token compression method for MLLM-based VQA tasks. QG-VTC employs a pretrained text encoder and a learnable feed-forward layer to embed user questions into the vision encoder's feature space then computes correlation scores between the question embeddings and visual tokens. By selecting the most relevant tokens and softly compressing others, QG-VTC ensures fine-tuned relevance to user needs. Additionally, a progressive strategy applies this compression across different vision encoder layers, gradually reducing token numbers. This approach maximizes retention of question-relevant information while discarding irrelevant details. Experimental results show that our method achieves performance on par with uncompressed models using just 1/8 of the visual tokens. The code and model will be publicly available on GitHub.

Xiaohui Li

One of the biggest challenges is to determine the deployment and navigation of the drones to benefit the most for different applications. Many research questions have been raised about this topic. For example, drone-enabled wildlife monitoring has received much attention in recent years. Unfortunately, this approach results in significant disturbance to different species of wild animals. Moreover, with the capability of rapidly moving communication supply towards demand when required, the drone equipped with a base station, i.e., drone-cell, is becoming a promising solution for providing cellular networks to victims and rescue teams in disaster-affected areas. However, few studies have investigated the optimal deployments of multiple drone-cells with limited backhaul communication distances. In addition, the use of autonomous drones as flying interactors for many real-life applications has not been sufficiently discussed. With superior maneuverability, drone-enabled autonomous aerial interacting can potentially be used on shark attack prevention and animal herding. Nevertheless, previous studies of autonomous drones have not dealt with such applications in much detail. This report explores the solutions to all the mentioned research questions, with a particular focus on the deployment and navigation of the drones. Simulations have been conducted to verify the effectiveness of the proposed approaches. We believe that our findings in this report shed new light on the fundamental benefits of autonomous civilian drones.

Xiaohui Li, Li Xing

Drone base stations can assist cellular networks in a variety of scenarios. To serve the maximum number of users in an area without apriori user distribution information, we proposed a two-stage algorithm to find the optimal deployment of drone base stations. The algorithm involves UTDOA positioning, coverage control and collision avoidance. To the best of our knowledge, the concept that uses network-based localization to optimize the deployment of drone-BSs has not been analyzed in the previous literature. Simulations are presented showing that the proposed algorithm outperforms random search algorithm in terms of the maximum number of severed users under the deployment of drone-BSs they found, with limited user densities.

Xiaohui Li

Drone base stations can provide cellular networks in areas that have lost coverage due to disasters. To serve the maximum number of users in the disaster area without apriori user distribution information, we proposed a 'sweep and search' algorithm to find the optimal deployment of drone base stations. The algorithm involves polygon area decomposition, coverage control and collision avoidance. To the best of our knowledge, this paper is the first in the literature that studied the deployment of drone base station without apriori user distribution information. Simulations are presented showing that the proposed algorithm outperforms the random search algorithm and the attractive search algorithm regarding the maximum number of severed users under the deployment of drone-BSs they found with a time limit.