3dr2n2-Deep-Learning-Course-project

Introduction

This repository contains the source codes for the paper Choy et al., 3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction, ECCV 2016. Given one or multiple views of an object, the network generates voxelized ( a voxel is the 3D equivalent of a pixel) reconstruction of the object in 3D.\

This repository is a part of Deep Learning Course Project which contains two phases. The first phase is to re-implement 3dr2n2 from scratch using tensorflow. The second phase is to improve the performance of 3d reconstructionddd.\

Datasets

We used ShapeNet models to generate rendered images and voxelized models which are available below: To download, use wget ftp://cs.stanford.edu/cs/cvgl/ShapeNetRendering.tgz.

• ShapeNet rendered images ftp://cs.stanford.edu/cs/cvgl/ShapeNetRendering.tgz
• ShapeNet voxelized models ftp://cs.stanford.edu/cs/cvgl/ShapeNetVox32.tgz

Training the network

• Download datasets and place them in a folder data
• Change dir to data_loader and run python save_files.py
• Change dir to mains and run python train.py