Technology Innovation Institute, Abu Dhabi, UAE
Dong-Wook Lee earned Ph. D in Texas Tech University and worked Purdue University as postdoc and Hangyang University as research assistant and associate professor and IMDEA as research associate and Masdar Institute of Science and Technology (later merged in Khalifa University) as research scientist before joining Technology Innovation Institute. He has expertise on several simulation techniques such as phase field method, finite element method and molecular dynamics for multiphysics and multiscale modeling.
We use rotational 2D STEM images to generate 3D STEM image and more STEM images are required to generate more accurate 3D STEM image, but it requires more time and effort. Sometimes, sample can be damaged when it is scanned so we have better to generate synthetic images from experimental images to save time and effort. Deep learning techniques is used to perform this task. Problem of next video frame prediction gets a lot of attention due to its application in many computer vision problems such as autonomous driving and structural health monitoring which has spatiotemporal images. Recently combinations of several deep learning models, such as autoencoder, recurrent neural network, convolution neural networks and long-short term memory, are developed to predict next frame based a serious of previous frames [1]. And frame interpolation, which synthetize intermediate images between a pair of input images, is also important in computer vision problem since it can be used for frame rate upsampling or creating slow-motion video effects. FILM model is developed for frame interpolation [2]. We use these two methods to reconstruct STEM images that are shown in the Figure 1 and 2. Frame interpolation method shows more accurate image than next frame prediction method. Images in blue box are real and in red box are synthetic images.
©khalifa university. All rights reserved.
Funded by
