Qimin Chen

I am a Research Assistant in Computer Science department at UC San Diego, advised by Prof. David J. Kriegman.

I received my M.S. degree in Computational Science from UC San Diego, where I worked at Center for Visual Computing Lab advised by Prof. David J. Kriegman. I also worked at Machine Learning, Perception, and Cognition Lab (mlPC) advised by Prof. Zhuowen Tu. Before that, I received my B.S degree in Computer Science and Technology from Fuzhou University.

qic003 at ucsd dot edu  /  CV  /  LinkedIn  /  Google Scholar  /  Github

I am also doing my summer research internship at the Brown Visual Computing group under Prof. Srinath Sridhar.
profile photo
Research

My research interest mainly focuses on Computer Vision/Graphics, especially geometric modeling as well as generative model for 3D shape and scene understanding and reconstruction. I am also interested in object detection and segmentation.

CoralNet Beta: Deploy Deep Neural Networks for Coral Reef Analysis

  • Developed core deep learning codebase for ResNet and EfficientNet efficient training and testing pipelines in Pytorch.
  • Worked on CoralNet deep learning backend development.
  • Evaluated image features extracted from CoralNet using Logistic Regression, Multi-Layer Perceptron, SVM, Random Forest and Gaussian Naive Bayes.

Topology-Aware Single-Image 3D Shape Reconstruction

  • Development of the topology-aware shape auto-encoder (TPWCoder) to address high-level topological properties such as genus and connectivity for 3D shape reconstruction.
  • Design of differentiable topological loss to combine with an end-to-end 3D reconstruction algorithm, MarrNet.
  • Noticeable qualitative and quantitative improvement over the state-of-the-art on the challenging ABC dataset.

Detecting Line Segments as Objects

  • Cast the line segment detection problem into an object detection task as formulating line segments as objects.
  • Proposed three options, Vanilla Box Diagonals, End-Point Shifts and Center Shift, Rotate and Scale, to formulate line objects to be end-to-end trainable.
Publications

Topology-Aware Single-Image 3D Shape Reconstruction
Qimin Chen, Vincent Nguyen, Feng Han, Raimondas Kiveris, Zhuowen Tu
CVPR Workshop on Learning 3D Generative Models (CVPRW), 2020
pdf / poster / bibtex

Composing volumetric-based generative model with topology-awareness auto-encoder allows them to learn high-level topological properties such as genus and connectivity for 3D shape reconstruction.

Teaching


Layout inspired by Jon Barron . Thank you Jon. © 2020 Deep Learning and Keep Learning.