Hi there, 👋
As a self-motivated software engineer, I have experience in Game Design, AI Design, Network
Routing, Cyber-security.
1 year SDE experience, Teaching Assistant in Algorithm and Data Structure, pursuing Master degree in Computer Science at Northeastern University.
Projects
No traditional pattern recognition method is used. Instead, by implementing neural network in pytorch package, pattern will be recognized through deep learning network. Here, pytorch is chosen since neural network could be customized easily in pytorch, whereas Keras provides a relatively high-level API or TensorFlow requires to setup everything to startup. Mnist digit recognition data set is primarily used for training and testing deep network in this project to avoid high requirement in computing power and to achieve the product easily. Analysis of first couples of convolution neural network layers and modified deep network will also be presented in the project.
An application to track predefined board, to use such board to find intrinsic and extrinsic parameters for camera calibration, and then to generate virtual objects with the right size and orientation in a scene according to the calibration of world coordinates and transformations (extrinsic parameters). Chessboard, Aruco board, or chessboard + Aruco board is selected as board because they are easy for computers to detect corners.
A real-time object detection system to capture 2D objects is built. Implemented tasks including thresholding from scratch, cleaning up from scratch by implementing grassfire, segmentation into the regions from scratch, computing features of each region, collecting training data, classifications by different classifiers, evaluating the performance, and a video demo. We also designed a user-friendly GUI by cvui . The following are instructions about our program.
A Regnet model with adabelief optimizer is built to detect fake face generated by StyleGAN 2 and SytleGAN 3 models.