Hello there! My name is Chi-Mao Fan, but feel free to call me Jonathan. I completed my master degree at National Chung Hsing University (NCHU), majoring in Computer Vision (CV) and Artificial Intelligence (AI).
Currently, I work as an R&D Engineer in the Digital Image Technology department at ASUS in Taiwan. I am extremely passionate about exploring new advancements in AI technology and am always eager to take on new and interesting projects.
If you would like to learn more about my experience and qualifications, please refer to the information below.
Master's degree in Electrical Engineering with a specialization in Computer Vision through Deep Learning and Artificial Intelligence, with a focus on Image Restoration. Achieved an outstanding GPA of 4.13/4.15 and participated in multiple deep learning competitions, including data analysis, object detection, and segmentation. Authored 4 conference papers, which were accepted by several international conference such as ICIP, EUSIPCO, and ISCAS, and received the 2022 IEEE Taipei Section Best Paper Awards for master thesis work. Passionate about using software skills to develop innovative solutions for real-world challenges. (You can check my publications as following)
Bachelor's degree in Electrical Engineering with a focus on Electronics, Circuitry, and Electromagnetism, as well as coursework in basic software courses such as data structures and C programming. Graduation project involved developing an Android app using JavaScript, SQL, and socket connections. Passionate about software development and leveraging programming skills to create functional and useful applications.
As an R&D Engineer at Digital Image Technology (DIT) in ASUS, I develop and implement computer vision algorithms using AI, while collaborating with both front-end and back-end teams to create innovative solutions for clients. My technical skills, communication abilities, and project management expertise have enabled me to successfully deliver projects and provide effective solutions to clients.
#Python #CSharp #C++ #OpenCV #PyTorch #Tensorflow #GitlabIn my second year of the master's program, I received an interview invitation from the Industrial Technology Research Institute's Information and Communications Labs, where I was responsible for supporting the Android system development team. I learned and practiced the technique of flashing Android image files onto development boards, enhancing my understanding of the Android operating system architecture and the application installation process.
#C #Android Development #Embedded Systems Compound Multi-branch Feature Fusion for Real Image Restoration
Image restoration is a challenging and ill-posed problem which also has been a long-standing issue. However, most of learning based restoration methods are proposed to target one degradation type which means they are lack of generalization. In this paper, we proposed a multi-branch restoration model inspired from the Human Visual System (i.e., Retinal Ganglion Cells) which can achieve multiple restoration tasks in a general framework. The experiments show that the proposed multi-branch architecture, called CMFNet, has competitive performance results on four datasets, including image deblurring, dehazing and deraindrop which are very common applications for autonomous cars.
Image Restoration using Improved Hierarchical Encoder-Decoder Networks
with Selective Residual Blocks
In this paper, we based on light hierarchical
network architecture: U-Net, and improve from Residual Dense Block (RDB)
which is good at image restoration tasks but memory-consuming to an efficient
block called Selective Residual Block (SRB). We also improve the hierarchical
network structure U-Net by adding the gatepost feature paths which enrich more
spatial feature information comparing with the traditional U-Net and have the
synergy with SRB. Besides this, we also proposed a loss function which is based
on two important metrics in image restoration: peak signal-to-noise (PSNR) and
structural similarity index to optimize our model. Finally, proposed network could
handle the 9 different restoration tasks such as denoising, deblurring, deraining,
dehazing and low-light image enhancement. Furthermore, the performances are
good in terms of quantitative metrics and visual quality. This paper was selected for the IEEE 2022 Taipei Section Best Master's and Ph.D. Thesis Award 🎉
Half Wavelet Attention on M-Net+ for Low-Light Image Enhancement
Low-Light Image Enhancement is a computer vision task which intensifies the dark images to appropriate brightness. It can also be seen as an illposed problem in image restoration domain. With the success of deep neural networks, the convolutional neural networks surpass the traditional algorithm-based methods and become the mainstream in the computer vision area. To advance the performance of enhancement algorithms, we propose an image enhancement network (HWMNet) based on an improved hierarchical model: M-Net+. Specifically, we use a half wavelet attention block on M-Net+ to enrich the features from wavelet domain. Furthermore, our HWMNet has competitive performance results on two image enhancement datasets in terms of quantitative metrics and visual quality.
Improved Hierarchical M-Net+ for Blind Image Denoising
Image denoising is a long standing ill-posed prob-lem. Recently, the convolution neural networks (CNNs) gradually stand in the spotlight and almost dominated the computer vision field and had achieved impressive results in different levels of vision tasks. One of famous hierarchical CNN-backbones is the U-Net which shows awesome performance in both denoising and other areas of computer vision. However, the hierarchical architecture usually suffers from the loss of spatial information due to the repeated sampling. It seriously affects the denoising performance especially the element-wise task like denoising. In this paper, we proposed an improved hierarchical backbone: M-Net+ for image denoising to ameliorate the loss of spatial details. Furthermore, we test on two synthetic Gaussian noise datasets to demonstrate the competitive result of our model.
Selective Residual M-Net for Real Image Denoising
Image restoration is a low-level vision task which is to restore degraded images to noise-free images. With the success of deep neural networks, the convolutional neural networks surpass the traditional restoration methods and become the main-stream in the computer vision area. To advance the performance of denoising algorithms, we propose a blind real image denoising network (SRMNet) by employing a hierarchical architecture improved from U-Net. Specifically, we use a selective kernel with residual block on the hierarchical structure called M-Net to enrich the multi-scale semantic information. Furthermore, our SRMNet has competitive performance results on two synthetic and two real-world noisy datasets in terms of quantitative metrics and visual quality. The source code and pretrained model are available at https://github.com/FanChiMao/SRMNet.
SUNet: Swin Transformer UNet for Image Denoising
Image restoration is a challenging ill-posed problem which also has been a long-standing issue. In the past few years, the convolution neural networks (CNNs) almost dominated the computer vision and had achieved considerable success in different levels of vision tasks including image restoration. However, recently the Swin Transformer-based model also shows impressive performance, even surpasses the CNN-based methods to become the state-of-the-art on high-level vision tasks. In this paper, we proposed a restoration model called SUNet which uses the Swin Transformer layer as our basic block and then is applied to UNet architecture for image denoising.
WBTP-VTON: Whole Body and Texture Preservation based Virtual Try-On Network
Image-based virtual clothes try-on systems are becoming more and more popular. However, many challenges are waiting to be solved. Therefore, we propose a new fully learnable method, called the whole body and texture preservation based virtual try-on network (WBTP-VTON) to guide the virtual attempt to deal with all practical challenges in this area. First, the WBTP-VTON template conversion is used to transform the target clothing and pants (or skirts) according to the body shape of the target person using a method called Geometric Matching Module (GMM). The second part is to synthesize the final image and make the generated results more realistic. Finally, we use try-on modules and synthetic masks to combine the deformed clothes and the final image to ensure image smoothness. After experimenting on a large data set, it is proved that our WBTP-VTON method has advanced virtual try-on performance.
