Jittor-GAN: A fast-training generative adversarial network model zoo based on Jittor

Background. The generation of medical images is to convert the existing medical images into one or more required medical images to reduce the time required for sample diagnosis and the radiation to the human body from multiple medical images taken. Therefore, the research on the generation of medical images has important clinical significance. At present, there are many methods in this field. For example, in the image generation process based on the fuzzy C-means (FCM) clustering method, due to the unique clustering idea of FCM, the images generated by this method are uncertain of the attribution of certain organizations. This will cause the details of the image to be unclear, and the resulting image quality is not high. With the development of the generative adversarial network (GAN) model, many improved methods based on the deep GAN model were born. Pix2Pix is a GAN model based on UNet. The core idea of this method is to use paired two types of medical images for deep neural network fitting, thereby generating high-quality images. The disadvantage is that the requirements for data are very strict, and the two types of medical images must be paired one by one. DualGAN model is a network model based on transfer learning. The model cuts the 3D image into multiple 2D slices, simulates each slice, and merges the generated results. The disadvantage is that every time an image is generated, bar-shaped “shadows” will be generated in the three-dimensional image. Method/Material. To solve the above problems and ensure the quality of image generation, this paper proposes a Dual3D&PatchGAN model based on transfer learning. Since Dual3D&PatchGAN is set based on transfer learning, there is no need for one-to-one paired data sets, only two types of medical image data sets are needed, which has important practical significance for applications. This model can eliminate the bar-shaped “shadows” produced by DualGAN’s generated images and can also perform two-way conversion of the two types of images. Results. From the multiple evaluation indicators of the experimental results, it can be analyzed that Dual3D&PatchGAN is more suitable for the generation of medical images than other models, and its generation effect is better.

Download Full-text

A Generative Adversarial Network Model for Simulating Various Types of Human-Induced Loads

International Journal of Structural Stability and Dynamics ◽

10.1142/s0219455419500925 ◽

2019 ◽

Vol 19 (08) ◽

pp. 1950092 ◽

Cited By ~ 3

Author(s):

Jiecheng Xiong ◽

Jun Chen

Keyword(s):

Network Model ◽

Generative Adversarial Networks ◽

Generative Adversarial Network ◽

Real Samples ◽

Load Model ◽

Adversarial Network ◽

Adversarial Networks ◽

Power Spectral ◽

Physical Features ◽

Fully Connected

Severe vibrations may occur on slender structures like footbridges and cantilever stands due to human-induced loads such as walking, jumping or bouncing. Currently, to develop a load model for structural design, the main features, such as periodicity and stationarity of experimental load records, are artificially extracted and then mathematically modeled. Different physical features have been included in different load models, i.e. no unified load model exists for different individual activities. The recently emerged generative adversarial networks can be used to model high-dimensional random variables. The probability distribution of these variables learned from real samples can be used to generate new samples, avoiding extracting features artificially. In this paper, a new model is proposed which combines the conditional generative adversarial networks and Wasserstein generative adversarial networks with gradient penalty to generate individual walking, jumping and bouncing loads. The generator of the model has five fully connected layers and a one-dimensional convolutional layer, and the discriminator has five fully connected layers. After one million training steps using the experimental records, the generator can generate high-quality samples similar to real samples in waveform. Finally, by comparing the power spectral densities and single degree of freedom system’s responses of the generated samples with real samples, it is further proved that the proposed generative adversarial network model can be used to simulate various human-induced loads. Source code of the model along with its trained weights is provided to the readers to further analysis and application.

Download Full-text

RSS Remeasurement Estimation for Indoor Positioning System with Generative Adversarial Network Model

2021 2nd International Conference on Computing, Networks and Internet of Things (CNIOT 2021) ◽

10.1145/3468691.3468742 ◽

2021 ◽

Author(s):

Xiaoqi Ren ◽

Wenyuan Tao ◽

Chung-Ming Own

Keyword(s):

Network Model ◽

Indoor Positioning ◽

Positioning System ◽

Generative Adversarial Network ◽

Adversarial Network ◽

Indoor Positioning System

Download Full-text

Perceptual Fusion of Electronic Chart and Marine Radar Image

Journal of Marine Science and Engineering ◽

10.3390/jmse9111245 ◽

2021 ◽

Vol 9 (11) ◽

pp. 1245

Author(s):

Chuang Zhang ◽

Meihan Fang ◽

Chunyu Yang ◽

Renhai Yu ◽

Tieshan Li

Keyword(s):

Low Frequency ◽

Wasserstein Distance ◽

Radar Image ◽

Navigation Systems ◽

Generative Adversarial Network ◽

Registration Method ◽

Noise Interference ◽

Adversarial Network ◽

Fast Training ◽

High Consistency

Electronic charts and marine radars are indispensable equipment in ship navigation systems, and the fusion display of these two parts ensures that the vessel can display dangerous moving targets and various obstacles on the sea. To reduce the noise interference caused by external factors and hardware, a novel radar image denoising algorithm using the concept of Generative Adversarial Network (GAN) using Wasserstein distance is proposed. GAN focuses on transferring the image noise distribution between strong and weak noise, while the perceptual loss approach is to suppress the noise by comparing the perceptual characteristics of the output after denoising. Afterwards, an image registration method based on image transformation is proposed to eliminate the imaging difference between the radar image and chart image, in which the visual attribute transfer approach is used to transform images. Finally, the sparse theory is used to process the high frequency and low frequency subband coefficients of the detection image obtained by the fast Fourier transform in parallel to realizing the image fusion. The results show that the fused contour has a high consistency, fast training speed and short registration time.

Download Full-text