【Academic Seminar】Smart 3D Shape Modeling and Reconstruction - Dr. Changjian Li

Topic: Smart 3D Shape Modeling and Reconstruction

Speaker: Dr. Changjian Li, The University of Hong Kong

Date and Time: 16:00 pm - 17:00 pm, Wednesday, September 25, 2019

Venue: Boardroom, Dao Yuan Building

Abstract：

Computer technology is stepping into our life, such as in entertainment, computer-aided medical treatment, manufacturing, and so on. Behind the success, 3D models are the key and the basic building block. Thus, how to efficiently acquire 3D models becomes the most challenge problem. To solve this, we conduct our research in two directions.

First, shape from user sketching. A novel approach, called BendSketch, is first presented which takes input user annotated sketches and predicts the freeform surface by exploiting geometric optimizations. Then a data-driven method, called SketchCNN is presented for modeling generic freeform 3D surfaces from sparse, expressive 2D sketches by incorporating convolutional neural network (CNN) into the sketch processing workflow. Both the non-learning and learning based methods have a great ability to model high-quality freefrom shapes making use of the proposed multi-view sketching pipeline. We validate our approaches, compare it with previous methods, and evaluate its performance with various modeling tasks. The results demonstrate our methods are new approaches for efficiently modeling freeform shapes with expressive 2D sketches.

Second, shape from images. We aim at reconstructing human teeth models from CBCT images and proposed ToothNet. The core of ToothNet is a two-stage network. In the first stage, an edge map is extracted from the input CBCT image to enhance image contrast along shape boundaries. In the second stage, we build our network upon the 3D region proposal network (RPN) with a novel learnedsimilarity matrix to help efficiently remove redundant proposals, speed up training and save GPU memory. To resolve the ambiguity in the identification task, we encode teeth spatial relationships as an additional feature input in the identification task, which helps to remarkably improve the identification accuracy. Our evaluation, comparison and comprehensive ablation studies demonstrate that our method produces accurate instance segmentation and identification results automatically and outperforms the state-of-the-art approaches.

Biography：

Dr. Changjian Li got his Ph.D. degree in the department of Computer Science, the University of Hong Kong, supervised by Prof. Wenping Wang. Before that, he got his Bachelor's degree from Shandong University in 2014. He has a great passion for working on Graphics and Vision research problems exploiting advanced deep learning techniques. He has a broad range of research interests, including machine learning and artificial intelligence, medical image processing, shape modeling, geometry processing, 3D vision. During his Ph.D. period, he has published several SIGGRAPH/SIGGRAPH Asia, CVPR, ICCV papers (see his personal page for details: https://enigma-li.github.io/).