Academy of Interdisciplinary Studies Division of Emerging Interdisciplinary Areas 200 Division of Emerging Interdisciplinary Areas Human Faces Generation using GAN Supervisor: HUI, Pan / EMIA Student: LYU, Zichen / COMP ZHANG, Chenyu / COMP Course: UROP1100, Fall UROP1100, Fall This UROP project is focusing on 3D human reconstruction using deep learning methods. A basic implementation of the topic is to construct the 3D model by single photo or multi-angle photos. Tasks assigned to us involved related paper reading, recurrence of existing models and presenting inspirations to the project. We will explain about the two papers we found most useful when on the reading list, as understandable a manner as possible with our own understanding as our final report. The first paper on RSCNet is about 3D human pose and shape estimation from low-resolution images; In the second paper, Neural 3D Mesh Renderer puts forward a method that first fitted polygon mesh into neural networks. They both provide us inspirations on our project. Human Faces Generation using GAN Supervisor: HUI, Pan / EMIA Student: LYU, Zichen / COMP WANG, Hesong / COMP ZHANG, Chenyu / COMP Course: UROP2100, Spring UROP1100, Spring UROP2100, Spring Our research project, conducted as part of the UROP 2100 program, aimed to explore recent advancement in 3D scene reconstruction and form representation. Our work can be divided into two parts: literature view and experimentation. For literature view, building on our previous focus on explicit representations and reconstruction in UROP 1100, we shifted our attention towards implicit representations during the this phase. Specifically, we studied several key articles, including ARCH, ARCH++, IDR, and DeepSDF, which helped broaden our understanding of this area. In the experimentation phase, we sought to reproduce the work of the previous semester, including ECON. Our main goal was to evaluate the reproducibility of the methods used in the earlier work and assess their effectiveness under different conditions. Overall, our project aimed to contribute to the ongoing research efforts in this exciting and rapidly evolving field of computer vision.
RkJQdWJsaXNoZXIy NDk5Njg=