The novel coronavirus (COVID-19) pandemic is rampant around the world, and teachers and students are unable to attend physical classes in the midst of a serious outbreak. This study aims to design a user-friendly, educational chatbot application interface that can be used as an after-school self-learning tool for students to enhance their interest and comprehension and increase the effectiveness of their learning at home. The system adopts the Chatfuel platform as the core interface and incorporates augmented reality technology to build a chatbot that allows users to interact with it after they have logged in to Facebook. The content is based on the biology subject of the first year of junior high school and is integrated into the online teaching with augmented reality teaching materials. A user survey is conducted to understand students’ attitudes towards learning biology with the aid of the ARCS motivation model, with 102 valid questionnaires received. The results show that the AR-based chatbot system developed in this study significant influenced the indicators in the ARCS motivation model; therefore, the intention to use the system is presumed to result in a noticeable increase in student learning outcomes when using the system. Accordingly, this study proposes new online learning tools for students to use at home during the pandemic, and the system also provides references for the future development and modification of educational chatbots.

Path planning is a key problem in the design of autonomous driving systems, and accurate traffic light detection is very important for robust routing. In this paper, we devise an object detection model, which mainly focuses on traffic light classification at a distance. In the past, most techniques employed in this field were dominated by high-intensity convolutional neural networks (CNN), and many advances have been achieved. However, the size of traffic lights may be small, and how to detect them accurately still deserves further study. In the object detection domain, the scheme of feature fusion and transformer-based methods have obtained good performance, showing their excellent feature extraction capability. Given this, we propose an object detection model combining both the pyramidal feature fusion and self-attention mechanism. Specifically, we use the backbone of the mainstream one-stage object detection model consisting of a parallel residual bi-fusion (PRB) feature pyramid network and attention modules, coupling with architectural tuning and optimizer selection. Our network architecture and module design aim to effectively derive useful features aimed at detecting small objects. Experimental results reveal that the proposed method exhibits noticeable improvement in many performance indicators: precision, recall, F1 score, and mAP, compared to the vanilla models. In consequence, the proposed method obtains good results in traffic light detection.

Due to the large volume of raw data in 3D point clouds, downsampling techniques are crucial for reducing computational load and memory usage to improve the training of 3D point cloud models. This paper plans to conduct research using the ModelNet40 dataset. Our proposed method is based on the PointNext architecture, an improved version of PointNet++ that significantly enhances performance through optimized training strategies and adjusted receptive fields. During the model training process, we employ the farthest point sampling method for downsampling. Specifically, we use an improved attention-based point cloud edge sampling (APES) method for downsampling, where we compute the density of each point and set the size of the neighbor K value to effectively retain feature points during downsampling. Our improved method captures edge points more effectively than the original APES method. By adjusting the architecture, our method, combined with the farthest point sampling method, not only reduced the average training time by nearly 15% compared to PointNext-s, but also improved accuracy from 93.11% to 93.57%.

As Virtual Try-On (VTON) technology matures, 2D VTON methods based on diffusion models can now rapidly generate diverse and high-quality try-on results. However, with rising user demands for realism and immersion, many applications are shifting towards 3D VTON, which offers superior geometric and spatial consistency. Existing 3D VTON approaches commonly face challenges such as barriers to practical deployment, substantial memory requirements, and cross-view inconsistencies. To address these issues, we propose an efficient 3D VTON framework with robust multi-view consistency, whose core design is to decouple the monolithic 3D editing task into a four-stage cascade as follows: (1) We first reconstruct an initial 3D scene using 3D Gaussian Splatting, integrating the SMPL-X model at this stage as a strong geometric prior. By computing a normal-map loss and a geometric consistency loss, we ensure the structural stability of the initial human model across different views. (2) We employ the lightweight CatVTON to generate 2D try-on images, that provide visual guidance for the subsequent personalized fine-tuning tasks. (3) To accurately represent garment details from all angles, we partition the 2D dataset into three subsets—front, side, and back—and train a dedicated LoRA module for each subset on a pre-trained diffusion model. This strategy effectively mitigates the issue of blurred details that can occur when a single model attempts to learn global features. (4) An iterative optimization process then uses the generated 2D VTON images and specialized LoRA modules to edit the 3DGS scene, achieving 360-degree free-viewpoint VTON results. All our experiments were conducted on a single consumer-grade GPU with 24 GB of memory, a significant reduction from the 32 GB or more typically required by previous studies under similar data and parameter settings. Our method balances quality and memory requirement, significantly lowering the adoption barrier for 3D VTON technology.

With Transformers achieving breakthrough results in natural language processing and computer vision, researchers have attempted to leverage their powerful modeling capabilities in 3D point cloud processing. However, the inherent unordered and unstructured nature of point cloud data poses significant challenges to directly applying Transformer architectures. This research proposes a novel point cloud processing method by introducing point cloud serialization and a serialization-based attention mechanism to enhance the performance of the PointNeXt model in semantic segmentation tasks. Traditional point cloud processing methods typically treat point clouds as unstructured data collections, resulting in low computational efficiency and scalability limitations. Our proposed approach breaks through the constraints of point cloud data’s unordered nature by serializing point clouds into a structured format. We employ spatial filling curves (such as Z-order and Hilbert curves) to sort point clouds, enabling efficient grouping of points into non-overlapping patches and applying more efficient attention mechanisms on these patches. Based on the serialization point cloud, we incorporate the segment attention mechanism from Point Transformer V3 (PTv3), which leverages the ordered characteristics of Serialization. By designing segment interactions (such as sequential shifting and sequential random mixing), we expand the model’s receptive field while maintaining computational efficiency.

Crop monitoring focuses on detecting and identifying numerous crops within a limited region. A major challenge arises from the fact that the target crops are typically smaller in size compared to the image resolution, as seen in the case of rice plants. For instance, a rice plant may only span a few dozen pixels in an aerial image that comprises thousands to millions of pixels. This size discrepancy hinders the performance of standard detection methods. To overcome this challenge, our proposed solution includes a testing time grid cropping method to reduce the scale gap between rice plants and aerial images, a multi-scale prediction method for improved detection using cropped images based on varying scales, and a mean-NMS to prevent the potential exclusion of promising detected objects during the NMS stage. Furthermore, we introduce an efficient object detector, the Enhanced CSL-YOLO, to expedite the detection process. In a comparative analysis with two advanced models based on the public test set of the AI CUP 2021, our method demonstrated superior performance, achieving notable 4.6% and 2.2% increases in F1 score, showcasing impressive results.