Dawei Guan

Mr. Dawei Guan

Beijing Jiaotong University, Beijing, China
orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
23 April 2025 in Sensors
Verified
Journal Article
Open access
PDF file
Dawei GuanDawei Guan
Dawei Guan
XLXinping Lin
Xinping Lin
HZHaoyi Zhang
Haoyi Zhang
HZHang Zhou
Hang Zhou
Image style transfer is a challenging task that has gained significant attention in recent years due to its growing complexity. Training is typically performed using paradigms offered by GAN-based image style transfer networks. Cycle-based training methods provide an approach for handling unpaired data. Nevertheless, achieving high transfer quality remains a challenge with these methods due to the simplicity of the employed network. The purpose of this research is to present VariGAN, a novel approach that incorporates three additional strategies to optimize GAN-based image style transfer: (1) Improving the quality of transferred images by utilizing an effective UNet generator network in conjunction with a context-related feature extraction module. (2) Optimizing the training process while reducing dependency on the generator through the use of a depthwise discriminator. (3) Introducing LPIPS loss to further refine the loss function and enhance the overall generation quality of the framework. Through a series of experiments, we demonstrate that the VariGAN backbone exhibits superior performance across diverse content and style domains. VariGAN improved class IoU by 236% and participant identification by 195% compared to CycleGAN.


14 December 2024
Verified
Conference Paper
Closed access
Dawei GuanDawei Guan
Dawei Guan
WLWei Li
Wei Li
Accurate and controllable image editing is a challenging task that has garnered significant attention in recent years. Pixel-level interactive editing is a hot research direction in this field. The point-based editing framework represented by DragGAN [1] proposes a standard paradigm for interactive editing. More recently, the combination of interactive editing with drag operations and the generative ability of diffusion models has led to even better solutions [2]. However, the generation effect of this scheme still suffers from issue of low robustness of the method. In this work, we proposed DenseDrag to optimize the diffusion based latent space editing process by introducing an additional technique: Dense-UNet latent space feature network. This method increases the semantic and information transmission ability of the latent space feature network, and improves the robustness of the method. We experimentally demonstrate the high quality of DenseDrag.

28 October 2024
Verified
Conference Paper
Closed access
YWYemeng Wu
Yemeng Wu
Dawei GuanDawei Guan
Dawei Guan
XLXinping Lin
Xinping Lin
ZQZifeng Qiu
Zifeng Qiu
HBHuihui Bai
Huihui Bai
Efficient and high-quality restoration of degraded images is a challenging task that has attracted much attention in recent years. Image restoration network based on Transformer provides a standard paradigm for network training process. Recently, better multi-scene image restoration methods have been developed based on Multi-head attention mechanism and larger receptive field. However, the restoration quality is still affected by the low robustness of the method. In this study, we introduce ClimaWeather, which enhances the Transformer block’s feature extraction by incorporating an efficient local feature extraction network, Local Enhance Feed Forward (LEFF) Block, and Select Kernel (SK) attention block. This method reduces the loss, improves the efficiency of feature extraction and the robustness of the model. We have demonstrated the superior performance of ClimaWeather through experiments.