Finding A Possible Solution for Shape From Shading Problem Using A Deep Learning Architecture.
Date of Submission
December 2020
Date of Award
Winter 12-12-2021
Institute Name (Publisher)
Indian Statistical Institute
Document Type
Master's Dissertation
Degree Name
Master of Technology
Subject Name
Computer Science
Department
Electronics and Communication Sciences Unit (ECSU-Kolkata)
Supervisor
Chanda, Bhabatosh (ECSU-Kolkata; ISI)
Abstract (Summary of the Work)
Given a single 2D image, our Objective is to pass that image into a Deep Learning Model and get the 3D shape of that image as output. These tasks have received many attentions recently, however, most of this existing approaches rely on 3D supervision, annotation of 2D images with keypoints, and also training with multiple views of each object instance, also known as Photometric stereo method. My frameworks are all based on the GAN architecture, with some custom made data set and also data from the ShapeNet data set, the models can run only on 2D images. Our model takes in a single image, encodes it into a latent code Z and passes it onto a generator, which the subsequently produces the depth image, which is used to construct the point cloud . The use of point cloud in the output representation, makes it possible for us to exploit the depth image of the object from the training image.
Control Number
ISI-DISS-2020-
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
DOI
NA
Recommended Citation
Sarkar, Soumyajit, "Finding A Possible Solution for Shape From Shading Problem Using A Deep Learning Architecture." (2021). Master’s Dissertations. 23.
https://digitalcommons.isical.ac.in/masters-dissertations/23
Comments
ProQuest Collection ID: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:28842732