ProjectsAll projects are open-sourced in my [Github]. I prefer to do something which really have an influence. And this page will be always updating. Research on rain and fog image clarity TechnologyThe project is based on the team's existing research results to achieve the task of rain and fog image clarity. In this project, my main work is model reproduction and training. At the same time, in order to increase the robustness of the model in the real scene, I am responsible for collecting some rain and fog data sets in the real scene and completing the data preprocessing. At present, the technology of this project has been supported by Tianjin fire brigade and China Fire Protection Association. In addition, the team also participated in the draft of the group standard of China Fire Protection Association "General requirements for virtual reality fire safety experience system" edited by Jiangxi Normal University. At present, this topic has been approved by the National College Students' innovation and entrepreneurship training program (Now the topic is finished, I am the project leader) Find out more details about this work. Self-supervised multi-scale pyramid fusion networks for realistic bokeh effect renderingThis work mainly produces a fuzzy aesthetic effect on the out of focus of a single image. By studying the top publications in the field, reproducing classical models, processing data sets and constantly conducting experiments. In August 2021, I proposed the smpnet algorithm of self supervised multi-scale out of focus scatter rendering. The main contribution is to introduce the idea of self-monitoring, build a multi-scale interactive fusion network of image pyramid, and the algorithm has achieved SOTA on related data sets. The article has been submitted to the SCI journal recommended by CCF: Journal of visual communication and image representation (Q2), which has been published. Find out more details about this work. A Dense Prediction ViT Network for Single Image Bokeh RenderingI apply vit to render scattered scenes, and further SOTA on some indicators. Relevant papers have been accepted by the 5th China pattern recognition and computer vision Conference (PRCV2022) Find out more details about this work. MCM/ICM Riders' riding strategies modelBased on the competition questions given by the organizer, I scientifically and reasonably formulated the riding strategies of cyclists. First, we divide the track into reasonable parts: flat road, uphill, downhill and curve. From the laws of kinematics and dynamics in physics, we can determine the basic handling rules of riders in different positions on the road, such as: flat road, constant speed driving; Uphill, accelerate in advance and slow down uphill; For long and gentle downhill, it is necessary to apply traction on the top of the slope to make it slide at a constant speed; For short and steep downhill, it can naturally accelerate and decline; For curves, slow down naturally when entering the curve and accelerate rapidly when leaving the curve. Based on the above rules, we build a power position dynamic model of riders. At the same time, we set the shortest goal, limiting the total energy that riders can consume, so that they will not be overloaded. |
