Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Tsung-Jung Liu has a total of 46 co-authors. Here are the top co-authors by number of shared publications:

  • Kuan-Hsien Liu

    National Taichung University of Science and Technology,Department of Computer Science and Information Engineering,Taichung,Taiwan,40401

    52 shared publications
  • Hsin-Hua Liu

    Graduate Institute of Communication Engineering, National Taiwan University

    16 shared publications
  • Soo-Chang Pei

    Graduate Institute of Communication Engineering, National Taiwan University

    14 shared publications
  • Chi-Mao Fan

    National Chung Hsing University,Department of Electrical Engineering and Graduate Institute of Communication Engineering,Taiwan

    9 shared publications
  • Kuan-Hsien Liu

    Department of Computer Science and Information Engineering, National Taichung University of Science and Technology, Taichung 40401, Taiwan

    7 shared publications
  • Weisi Lin

    School of Computer Engineering, Nanyang Technological University, Singapore 639798

    4 shared publications

Latest Publications

  • Open Access

    Face Aging Synthesis by Deep Cycle Generative Adversarial Networks and Bias Loss

    by Tsung-Jung Liu, Chia-Ching Wang

    In this work, we propose a deep learning-based adversarial network for human face aging synthesis. We use an image dataset consisting of young (Domain X) and elderly (Domain Y) individuals to generate aged images at various target ages. Our network employs two adversarial components: a generator that synthesizes aged images and a discriminator that verifies their authenticity. Through adversarial training, the generator iteratively improves its output based on feedback from the discriminator, ultimately achieving a Nash equilibrium where the generated images closely match the target ages. Moreover, we integrate deep ResNet blocks in the generator, with the best image quality achieved using a 7-layer configuration. To address artifacts caused by overly strict discriminators, we introduce a novel bias loss function that relaxes the discriminator’s harshness, resulting in more realistic aging effects. Additionally, we incorporate perceptual loss using a pre-trained VGG16 network to preserve identity features during age progression. Experimental results show that the proposed architecture significantly enhances aging realism and identity preservation, outperforming existing models both qualitatively and quantitatively. The source code and pre-trained models are available at https://github.com/Tony00728/FAS-by-CGAN-and-BiasLoss.


  • RDLNET: Residual Dense Block based Lightweight Network for Video Super-Resolution

    by Kuan-Hsien Liu, Chih-Jung Wang, Tsung-Jung Liu, Wen-Ren Liu

    Deep learning has been widely used in video super-resolution (VSR). Most VSR methods focus on achieving better quality, and the design of deep neural networks is also becoming more complex. Since the input data of video super-resolution is already huge, if the neural network is too complex, it will cause a higher memory load, and a higher-end graphic card device is required. In order to reduce the cost of VSR system, in the paper, we propose a new method called RDLNET, which is a residual dense block based neural network with lightweights to deal with VSR. In the experiments, our proposed lightweight VSR method reduces 25% parameters and maintains almost the same PSNR compared with other state-of-the-art VSR methods. The source code of RDLNET is made available at GitHub https://github.com/nutcliu2507/RDLNET.


  • LIRSRN: A Lightweight Infrared Image Super-Resolution Network

    by Chun-An Lin, Tsung-Jung Liu, Kuan-Hsien Liu

    Infrared imaging is crucial for many applications, such as night vision and environmental monitoring. However, it often suffers from lower spatial resolution compared to visible light imaging, which hinders its effectiveness in scenes demanding fine detail discernment. This paper introduces a Lightweight Infrared Image Super-Resolution Network (LIRSRN), which is a novel architecture designed to enhance the resolution of IR images with minimal computation overhead. The core of LIRSRN is the Attention Enhancement Module, which synergizes various attention mechanisms to emphasize important features across both channel and spatial domains. It is further enhanced by spatial frequency processing, which helps to extract key image attributes. The training is conducted on the DIV2K dataset, and the test results from "results-A" and "results-C" demonstrate the superior performance of LIRSRN in terms of PSNR and SSIM. Ablation studies compare the trade-offs between performance and computational cost using different attention mechanisms. Overall, the proposed model strikes a balance between performance and complexity. The source code and trained model are available at https://reurl.cc/yYDnA6.


  • SOAda-YOLOR: Small Object Adaptive YOLOR Algorithm for Road Object Detection

    by Yu-Fang Huang, Tsung-Jung Liu, Chun-An Lin, Kuan-Hsien Liu

    Nowadays, road object detection has become a challenging topic and a long-standing problem. In previous algorithms, traditional convolutional neural networks (CNNs) have almost dominated computer vision and achieved considerable successes in different detection tasks. However, Transformer-based models in the past two years have shown impressive performance, and some of which even surpass CNN-based methods and become the state-of-the-art for advanced vision tasks. In this paper, we propose a detection model named SOAda-YOLOR, which uses the current mainstream Bi-FPN concept with the innovative Optimized-NMS. We also propose a novel ShuffleSimAM module to the backbone, and apply some bag-of-freebies such as SIoU loss to road object detection, and finally achieve competitive results.


  • Image Inpainting by Mscswin Transformer Adversarial Autoencoder

    by Bo-Wei Chen, Tsung-Jung Liu, Kuan-Hsien Liu

    Image inpainting has been researched for years. From deeper and larger models to models that focus on global information, all of them aim to obtain results closer to reality. In this paper, we combine the stripe window and line-by-line feature shift to modify the Vision Transformer (ViT) to reduce the computation cost and obtain global information from the oblique attention. In addition, we design a new loss function to enhance the texture and colors for inpainting. At last, to validate the efficacy of our proposed model, we conduct extensive experiments on commonly seen datasets (Places2 and CelebA) compared with other state-of-the-art methods. The source code and pretrained models are available at https: //github.com/bobo0303/MSCS-Net.


  • Compound Multi-Branch Feature Fusion for Image Deraindrop

    by Chi-Mao Fan, Tsung-Jung Liu, Kuan-Hsien Liu

    Image restoration is a challenging and ill-posed problem which also has been a long-standing issue. In this paper, we proposed a multi-branch restoration model inspired from the Human Visual System (i.e., Retinal Ganglion Cells) for image deraindrop. The experiments show that the proposed multi-branch architecture, called CMFNet, has state-of-the-art performance results. The source code and pretrained models are available at https://github.com/FanChiMao/CMFNet. And the interactive demonstration of the proposed deraindrop model can be accessed at https://reurl.cc/dXaeNg.


  • Facial Expression Recognition in the Wild Using FAM, RRDB and Vision Transformer Based Convolutional Neural Networks

    by Kuan-Hsien Liu, Xiang-Kun Shih, Tsung-Jung Liu, Wen-Ren Liu

    Facial expression recognition in the wild is a quite challenge task because facial images captured in natural settings are often affected by various factors such as irregular occlusions, inconsistent face angles and varying light levels. To overcome these factors, we presented a new triple-branch deep neural network model containing facial attention module, residual in residual dense block and vision transformer to deal with facial expression recognition problem. The facial attention module can help backbone network extract more useful features. Residual in residual dense block and vision transformer can further get more detailed features. In addition to the image problems caused by above factors, the ratio of number of images for different expressions in the dataset is also very different. We proposed a new loss function to tackle this uneven ration problem. Experimental results on two benchmark in-the-wild datasets show that our model is indeed helpful for images in the wild. We also created a new expressions dataset called Fairfaceplus, which is built by adding expression categories to the original label on FairFace dataset. The code of our proposed method will be made available on GitHub https://github.com/dreampledge/fairfaceplus.


  • Scale Selection Network with Attention Mechanism for Crowd Counting

    by Ting-Hsu Lai, Tsung-Jung Liu, Kuan-Hsien Liu

    Crowd counting is an important cornerstone in tasks related to crowd analysis. In recent years, there has been a significant rise in the development of deep learning methods, particularly those based on Convolutional Neural Networks (CNN), in the field of image processing and achieved amazing results. However, there are some challenges waiting to be solved in the current crowd counting task: large-scale variations in crowd sizes and interference from the background. Both of them will lead to poor prediction results. Therefore, we propose a scale selection module to deal with the scale variation problem in images. And for background interference, we proposed an attention module to reduce the interference of background information. Moreover, we evaluated our model on four commonly used datasets and compared the performance with other state-of-the-art methods to demonstrate the competitiveness of our approach. The source code and pretrained models are available at https://git;hub.com/TimLai0307/SSN.


  • Open Access

    Lightweight Image Inpainting by Stripe Window Transformer with Joint Attention to CNN

    by Tsung-Jung Liu, Bo-Wei Chen, Kuan-Hsien Liu

    Image inpainting is an important task in computer vision. As admirable methods are presented, the inpainted image is getting closer to reality. However, the result is still not good enough in the reconstructed texture and structure based on human vision. Although recent advances in computer hardware have enabled the development of larger and more complex models, there is still a need for lightweight models that can be used by individuals and small-sized institutions. Therefore, we propose a lightweight model that combines a specialized transformer with a traditional convolutional neural network (CNN). Furthermore, we have noticed most researchers only consider three primary colors (RGB) in inpainted images, but we think this is not enough. So we propose a new loss function to intensify color details. Extensive experiments on commonly seen datasets (Places2 and CelebA) validate the efficacy of our proposed model compared with other state-of-the-art methods. Index Terms: HSV color space, image inpainting, joint attention, stripe window, transformer


  • Facial Expression Recognition in the Wild from Attention to Vision Transformer based CNNs

    by Kuan-Hsien Liu, Xiang-Kun Shih, Tsung-Jung Liu

    Facial expression recognition in the wild could be affected by irregular occlusions, inconsistent face angles and varying light levels on facial images. In this work, we proposed a model containing facial attention module, residual in residual dense block and vision transformer to tackle with this problem. The facial attention module can assist the model to extract more useful features. Residual in residual dense block and vision transformer can help the model extract more detailed features. In addition to, the number of images of different facial expressions in the dataset is also very different. We proposed a new loss function to deal with this issue. Experimental results on two in-the-wild datasets show that our model is indeed effective. We also created a new facial expression dataset called Fairfaceplus, which is an extension of FairFace, with annotation of expression categories to the original labels. The code of our proposed method will be available on GitHub https://github.com/dreampledge/fairfaceplus.