Landmark identification is crucial in quantifying cephalometric analysis such as Steiner, Bjork, Ricketts, Kim, Nagasaki, etc. For applying these analysis to cephalometric image, dentist or orthodontist need to annotate some group of landmarks that corresponding to specified analysis.
Cervical vertebra maturation (CVM) staging in lateral cephalometric radiographs is an efficient method to determine skeletal maturation, as lateral cephalometric radiography is routinely required for orthodontic diagnosis and treatment planning in orthodontic practice with no additional radiographs required to assess the CVM stages.
In this project, we focus on building-up an Instance segmentation model for recognizing drawing room layout data. We trained and tested our model (Unet architecture) on both public dataset (Cubicasa, R3D) and own dataset (more than 1000 images of drawing layout) To ensuring generalization of the trained model, we added CVF-FP dataset (public) to test set for evaluation.
In this project, we focus on synthesizing 360 panoramic image data from 3D models without real image. We first collected 3D models of in-house objects such as table, curtain, chairs, air-conditioner, TV, etc from online storage.
In this project, we focus on optimizing video object tracking and segmentation model for running on digital still camera. We used the Siamese network based model for tracking object on input video.
Researching some (xAI methods) for time series data. Eg: video action recognition. Implementing GradCAM, Contrastive Layer-wise Relevance Propagation (CLRP), Deep Taylor Decomposition algorithms for visualizing heat map (target feature map) that pretrained deep learning model focused on.