Generative adversarial network (GAN)-based data augmentation is used to enhance the performance of object detection models. It comprises two stages: training the GAN generator to learn the distribution of a small target dataset, and sampling data from the trained generator to enhance model performance. In this paper, we propose a pipelined model, called robust data augmentation GAN (RDAGAN), that aims to augment small datasets used for object detection. First, clean images and a small datasets containing images from various domains are input into the RDAGAN, which then generates images that are similar to those in the input dataset. Thereafter, it divides the image generation task into two networks: an object generation network and image translation network. The object generation network generates images of the objects located within the bounding boxes of the input dataset and the image translation network merges these images with clean images. A quantitative experiment confirmed that the generated images improve the YOLOv5 model’s fire detection performance. A comparative evaluation showed that RDAGAN can maintain the background information of input images and localize the object generation location. Moreover, ablation studies demonstrated that all components and objects included in the RDAGAN play pivotal roles.

The viewing time of media content per week through TV is still dominant. Users are exposed to numerous advertisements, such as commercials, electronic home shopping, product placement (PPL), and T-Commerce while watching TV programs. Most of the advertisement systems provide a good overview of products. However, traditional advertising services do not consider user preferences, meaning it is difficult to expect anything more than mere exposure to them. We can adopt a recommendation system to predict the preference. However, existing recommendation systems find it difficult to satisfy the real-time requirements of online broadcasting because of the large overhead incurred in preference prediction processes. In this paper, we propose a real-time recommendation system to provide personalized advertisements. The proposed system generates tree models based on user historical data. To reduce the overhead of preference prediction, we introduce a sorted HashMap that enables fast tree searches. For sophisticated preference prediction, the proposed system normalizes the users’ preferences by considering the characteristics of their tree model. Finally, we conduct experiments to evaluate the performance of the proposed tree-based recommendation system.

Object detection is a significant activity in computer vision, and various approaches have been proposed to detect varied objects using deep neural networks (DNNs). However, because DNNs are computation-intensive, it is difficult to apply them to resource-constrained devices. Here, we propose an on-device object detection method using domain-specific models. In the proposed method, we define object of interest (OOI) groups that contain objects with a high frequency of appearance in specific domains. Compared with the existing DNN model, the layers of the domain-specific models are shallower and narrower, reducing the number of trainable parameters; thus, speeding up the object detection. To ensure a lightweight network design, we combine various network structures to obtain the best-performing lightweight detection model. The experimental results reveal that the size of the proposed lightweight model is 21.7 MB, which is 91.35% and 36.98% smaller than those of YOLOv3-SPP and Tiny-YOLO, respectively. The f-measure achieved on the MS COCO 2017 dataset were 18.3%, 11.9% and 20.3% higher than those of YOLOv3-SPP, Tiny-YOLO and YOLO-Nano, respectively. The results demonstrated that the lightweight model achieved higher efficiency and better performance on non-GPU devices, such as mobile devices and embedded boards, than conventional models.

In the current era of online information overload, recommendation systems are very useful for helping users locate content that may be of interest to them. A personalized recommendation system presents content based on information such as a user’s browsing history and the videos watched. However, information filtering-based recommendation systems are vulnerable to data sparsity and cold-start problems. Additionally, existing recommendation systems suffer from the large overhead incurred in learning regression models used for preference prediction or in selecting groups of similar users. In this study, we propose a preference-tree-based real-time recommendation system that uses various tree models to predict user preferences with a fast runtime. The proposed system predicts preferences based on two balance constants and one similarity threshold to recommend content with a high accuracy while balancing generalized and personalized preferences. The results of comparative experiments and ablation studies confirm that the proposed system can accurately recommend content to users. Specifically, we confirmed that the accuracy and novelty of the recommended content were, respectively, improved by 12.1% and 27.2% compared to existing systems. Furthermore, we verified that the proposed system satisfies real-time requirements and mitigates both cold-start and overfitting problems.

A deep neural network (DNN) has become a key technique in many intelligent application domains. Since the DNN model has dozens of layers and millions of levels of parameters, the machine has to execute computation-intensive workloads. Therefore, it is difficult to perform DNN inference in resource-constrained devices such as mobile devices. In this paper, we propose a context-aware model selection for on-device DNN inference. The proposed model selection chooses a DNN model corresponding to a spatiotemporal domain based on the context information of the device. Since a context-aware model detects related objects by spatiotemporal domain, it has a low dimension of parameters. In resource-constrained environments, the proposed context-aware model enables high-accuracy inference at low latency. To evaluate the performance of the proposed model selection, we conduct comparison experiments with the existing object detection model. Through experiments, we confirm that the context-aware model performs better than the existing trained models when on-device object detection is performed. Finally, we discuss the limits of the proposed model selection.