Research Project
- Japan Urban Road Marking Segmentation
This project focuses on the problem of the existence of objects at multiple scales of the Road Marking Dataset, and investigates four kinds of network architectures that deal with the multiscale context. Inspired by previous studies, we propose a novel multi scale attention based dilated CNN to tackle the RMD. An attention module that can softly weight the feature maps from different scales and dilated convolution to enlarge the receptive field of feature maps (utilize a large range of spatial-context information) are adopted. The two models that employ scales=⟨0.5,1.0⟩(Model 1) and scales=⟨1.0,2.0⟩(Model 2) are trained to evaluate five kinds of multiscale inputs on the RMD. At the inference time, Model 2, with scales={0.5,1.0,2.0}, gained the best mIoU of 74.88%. The ablation study shows that the proposed method yields the best results by combining multiscale attention and dilated convolution.
- Mapping Road Marking Quality at lane level and city scale
Ongoing
- Attenuation Relationship of Peak Ground Velocity
The attenuation relationship of PGV is a method to predict the peak ground velocity of earthquake that may occur in the future based on the ground motion records of past earthquakes. The attenuation refers to the phenomenon that the farther away from the epicenter, the weaker the earthquake intensity. The attenuation relationships are used in both deterministic and probabilistic seismic hazard analyses. The previous attenuation relationships are empirical equations that predict the level of ground shaking, based on the source characteristics (e.g., earthquake magnitude), the propagation path (e.g., the shortest distance from the fault), and the local site conditions, etc. As the development of statistical analysis methods and more ground motion records are obtained, the research of attenuation relationship has been greatly developed. However, due to the lack of ground motion records near the epicenter, it was found that previous attenuation relationships have low reliability at close range. Therefore, it is useful to develop new attenuation relationships of peak ground velocity (PGV) using machine learning methods. This project tries to develop new attenuation relationships of peak ground velocity using machine learning methods: random forest, neural network, support vector machine, and XGBoost. In order to compare with the predictors obtained by machine learning, we have also constructed a new attenuation relationship of peak ground velocity using three-stage regression procedure.
Dataset Project
- RMD(Road Marking Dataset)
- RMQD(Road Marking Quality Dataset)
Ongoing
Playground Project
- Deep Learning based MongolGer&RomanTent Classification application from scratch
Ongoing
- Folium
Ongoing