Showing papers by "Tran Quang Trung published in 2022"
TL;DR: A novel deep neural network architecture called BlazeNeo is introduced, for the task of polyp segmentation and neoplasm detection with an emphasis on compactness and speed while maintaining high accuracy, and achieves improvements in latency and model size while maintaining comparable accuracy against state-of-the-art methods.
Abstract: In recent years, computer-aided automatic polyp segmentation and neoplasm detection have been an emerging topic in medical image analysis, providing valuable support to colonoscopy procedures. Attentions have been paid to improving the accuracy of polyp detection and segmentation. However, not much focus has been given to latency and throughput for performing these tasks on dedicated devices, which can be crucial for practical applications. This paper introduces a novel deep neural network architecture called BlazeNeo, for the task of polyp segmentation and neoplasm detection with an emphasis on compactness and speed while maintaining high accuracy. The model leverages the highly efficient HarDNet backbone alongside lightweight Receptive Field Blocks and a feature aggregation mechanism for computational efficiency. An auxiliary training strategy is proposed to take full advantage of the training data for the segmentation quality. Our experiments on a challenging dataset show that BlazeNeo achieves improvements in latency and model size while maintaining comparable accuracy against state-of-the-art methods. We obtain over 155 fps while outperforming all compared models in terms of accuracy in INT8 precision when deploying on a dedicated edge device with a conventional configuration.
11 citations
01 Oct 2022
TL;DR: Comparison results show that the proposed method to increase the resolution of thermal images using edge features of corresponding high-resolution visible images has the highest performance in terms of PSNR and SSIM.
Abstract: Thermal imaging has played an important role in a wide range of areas of life. However, thermal cameras often produce low-resolution images, which limits the ability to observe objects in thermal imaging applications. Modern thermal cameras often include a built-in high-resolution visible camera to supplement the thermal image information. This paper proposes a method to increase the resolution of thermal images using edge features of corresponding high-resolution visible images. The Canny edge detection and thin-line downscaling algorithms are used to generate edge maps from high-resolution visible images to contribute to the super-resolution network. The proposed super-resolution model is designed based on generative adversarial network architecture for $\times$ 2, $\times$ 3, and $\times$ 4 magnification. The KAIST dataset is used to train and test the model. Peak signal-to-noise ratio (PSNR) and structure similarity index (SSIM) are used to evaluate the quality of super-resolution images. After the training process, to prove the effectiveness of edge features, we compare the quality of the super-resolution images generated from the proposed method with other methods. The comparison results show that the proposed method has the highest performance in terms of PSNR and SSIM.
2 citations
TL;DR: In this article , a stretchable broadband photodetectors (PDs) were fabricated by forming organic-inorganic vertical multiheterojunctions on a three-dimensional micro-patterned stretchable substrate (3D-MPSS).
Abstract: Stretchable broadband photodetectors (PDs) are attractive for applications in wearable optoelectronics and personal healthcare. However, the development of stretchable broadband PDs is limited by difficulties in obtaining materials, designing device structures, and finding reliable fabrication processes. Here, we report stretchable broadband PDs by forming organic-inorganic vertical multiheterojunctions on a three-dimensionally micro-patterned stretchable substrate (3D-MPSS). The stress-adaptable 3D-MPSS structure allows all layers of the PD coated on it to sustain tensile strains. Generation of photovoltage in the vertical hybrid structure of PbS quantum dots/ZnO nanorods as a photo-responsive material on poly(3,4-ethylenedioxythiophene) poly(styrenesulfonate) as a transport channel is considred to be the mechanism of the device response to UV-Vis-NIR. The fabricated PDs present responsivity to UV (365 nm), Vis (565 nm and 660 nm), and NIR (880 nm and 970 nm) light, as well as reliable electrical performance under applied stretching up to 30%.
1 citations