Viet-Tham Huynh

Sketch classification plays a crucial role across diverse domains, including image retrieval, artistic style analysis, and content-based image retrieval. While CNNs have demonstrated remarkable success in various image-related tasks, the computational complexity of large models poses challenges in resource-constrained environments. To address this concern, we propose MobileNet-SA, a novel lightweight model that seamlessly integrates a self-attention module into the MobileNet architecture, with a specific focus on enhancing sketch classification performance. The MobileNet-SA model leverages the inherent efficiency of lightweight CNN while harnessing the power of self-attention mechanisms to effectively capture spatial dependencies and enrich feature representations within sketch data. In our experiments, MobileNet-SA achieves state-of-the-art results, demonstrating an impressive accuracy of 93.5% on the challenging SketchyCOCO dataset and 96.7% on the GM-Sketch dataset. We thoroughly evaluate the model’s performance across diverse sketch classes, confirming its robustness and generalization capabilities, which make it well-suited for real-world applications where input sketches may exhibit significant variations. Our research indicates that MobileNet-SA not only outperforms existing methods but also offers an efficient and interpretable solution for sketch classification tasks.

Information retrieval is vital in our daily lives, with applications ranging from job searches to academic research. In today’s data-driven world, efficient and accurate retrieval systems are crucial. Our research focuses on video data, using a system called LUMOS-DM: Landscape-based Multimodal Scene Retrieval Enhanced by Diffusion Model. This system leverages Vision Transformer and Diffusion Models, taking user-generated sketch images and text queries as input to generate images for video retrieval. Initial testing on a dataset of 100 h of global landscape videos achieved an 18.78% at Top-20 accuracy rate and 36.45% at Top-100 accuracy rate. Additionally, video retrieval has various applications, including generating data for advertising and marketing. We use a multi-modal approach, combining sketch and text descriptions to enhance video content retrieval, catering to a wide range of user needs.

Big cities are well-known for their traffic congestion and high density of vehicles such as cars, buses, trucks, and even a swarm of motorbikes that overwhelm city streets. Large-scale development projects have exacerbated urban conditions, making traffic congestion more severe. In this paper, we proposed a data-driven city traffic planning simulator. In particular, we make use of the city camera system for traffic analysis. It seeks to recognize the traffic vehicles and traffic flows, with reduced intervention from monitoring staff. Then, we develop a city traffic planning simulator upon the analyzed traffic data. The simulator is used to support metropolitan transportation planning. Our experimental findings address traffic planning challenges and the innovative technical solutions needed to solve them in big cities.

Recognizing hand-drawn sketches is a promising starting point for various applications, such as assisting artists in creating 3D environments for games or virtual environment scenes quickly and efficiently from concept arts. In addition, by understanding drawings, we can generate 3D models that can be used for further design and development. Thus, in this paper, we aim to develop a novel lightweight network that can accurately recognize sketch drawings. We propose a lightweight-yet-efficient neural network based on MobileNetV2 for sketch recognition and employ knowledge distillation to train the proposed model from EfficientNet-B4. To evaluate the accuracy of the proposed method, we collect a dataset of sketches comprising 1800 drawings in 12 categories, ranging from furniture to animals. The experimental results show that our network model achieves an accuracy of 96.7%, with 96.9% precision, 96.7% recall, and 96.7% F1-score. These results demonstrate that the proposed approach has great potential for practical sketch recognition applications, such as interior design or VR scene generation.