THREE DIMENSIONAL RECONSTRUCTION FROM SINGLE VIEW TWO DIMENSIONAL IMAGE

• Main Author: Timothy Panjaitan, David
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/48221
• Institution: Institut Teknologi Bandung

Methods for the reconstruction of three-dimensional models from a single view twodimensional image have developed rapidly in recent years. However, most of these methods are still in the form of research with a very controlled form of input. For example, in the Pixelaligned Implicit Function method, reconstruction has succeeded in achieving a threedimensional model with highly detailed human shapes and textures from a single view twodimensional image, but the input image is limited by various criteria and must include a mask image that shows the human position in the image. Therefore, this final project aims to build applications that can reconstruct a three-dimensional model of a single view two-dimensional image with broader criteria and do not require additional input other than the relevant twodimensional picture. For this reason, a segmentation module will be created that uses a semantic segmentation method named Fully Convolutonal DenseNet to produce the mask image needed for reconstruction. This segmentation module will also allow for handling object reconstruction for images containing more than one object. The use of segmentation modules results in a decrease in the quality of reconstruction but it also increases automation and enlarges the range of inputs that can be processed.