AVS1508/COMPSCI-591NR-Project

COMPSCI 591NR Project - Spring 2023

Topic: 3D Scene/Object Reconstruction using CycleGANs with Binocular Stereopsis

Group Members

• Aditya Vikram Singh (avsingh@umass.edu)
• Robert Washbourne (rwashbourne@umass.edu)
• Rohan Raghuram (rraghuram@umass.edu)

Generating Stereoscopic Renders via Depth Renderer on ShapeNet subset

Object Class	Stereo Pair 1 Left Image	Stereo Pair 1 Right Image	Stereo Pair 2 Left Image	Stereo Pair 2 Right Image
Bed 1
Bed 2
Chair 1
Chair 2

Generating Depth/Disparity Map from Pair of Steroscopic Images

Input Image (Left)	Input Image (Right)	Generated Depth Map	RGB-SD Image

References and Works Used

1. Depth_Renderer

   @inproceedings{
       NEURIPS2020_ba036d22,
       author = {Nie, Yinyu and Lin, Yiqun and Han, Xiaoguang and Guo, Shihui and Chang, Jian and Cui, Shuguang and Zhang, Jian.J},
       booktitle = {Advances in Neural Information Processing Systems},
       editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},
       pages = {16119--16130},
       publisher = {Curran Associates, Inc.},
       title = {Skeleton-bridged Point Completion: From Global Inference to Local Adjustment},
       url = {https://proceedings.neurips.cc/paper/2020/file/ba036d228858d76fb89189853a5503bd-Paper.pdf},
       volume = {33},
       year = {2020}
   }

2. GenRen

   @inproceedings{
       9320324,
       author={Aumentado-Armstrong, Tristan and Levinshtein, Alex and Tsogkas, Stavros and Derpanis, Konstantinos G. and Jepson, Allan D.},
       booktitle={2020 International Conference on 3D Vision (3DV)}, 
       title={Cycle-Consistent Generative Rendering for 2D-3D Modality Translation}, 
       year={2020},
       volume={},
       number={},
       pages={230-240},
       doi={10.1109/3DV50981.2020.00033}
       }