Skip to content
/ oasis Public
forked from princeton-vl/oasis

Code for the CVPR 2020 paper "OASIS: A Large-Scale Dataset for Single Image 3D in the Wild"

License

Notifications You must be signed in to change notification settings

dfan/oasis

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OASIS: A Large-Scale Dataset for Single Image 3D in the Wild

OASIS

This repository contains the code for the following paper:

OASIS: A Large-Scale Dataset for Single Image 3D in the Wild,
Weifeng Chen, Shengyi Qian, David Fan, Noriyuki Kojima, Max Hamilton, Jia Deng
Conference on Computer Vision and Pattern Recognition (CVPR), 2020.

Please check the project site for more details.

Installation

The code has been tested on python 3.7, cuda 10.0, pytorch 1.1.0, gcc 8.4.0

conda create --name oasis python=3.7
conda activate oasis

conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=10.0 -c pytorch  
conda install opencv==3.4.2 h5py scipy pillow==6.1.0 scikit-learn
pip install sacred easydict pyyaml imageio==2.6.0 tb-nightly future tqdm

Data Preparation

Please go to the download page and download all the images and annotations. Then untar:

mkdir OASIS
tar -xzf OASIS_images_v1.tar.gz  -C OASIS
tar -xzf OASIS_trainval_annotations_v1.tar.gz -C OASIS

The folder tree after these steps should look like:

OASIS
    - LICENSE
    - OASIS_trainval
        - image
        - meta
        - OASIS_train.csv
        - OASIS_val.csv
        - depth
        - normal
        - fold
        - occlusion
        - mask
        - DIW_style_rel_depth
        - segmentation
            - planar_instance
            - continuous_instance
    - OASIS_test
        - image
        - meta
        - OASIS_test.csv    

Experiments

The experiment folder contains code to reproduce the results for the following experiments:

  • Depth Estimation
  • Surface Normal Estimation
  • Fold and Occlusion Boundary Detection
  • Planar Instance Segmentation

Please refer to the README files under each folder for instructions on how to run the code.

To run on pretrained models, please first download the pretrained models experiments.tar.gz, and tar -xzf experiments.tar.gz.

About

Code for the CVPR 2020 paper "OASIS: A Large-Scale Dataset for Single Image 3D in the Wild"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • MATLAB 40.1%
  • HTML 31.8%
  • Python 18.8%
  • C++ 7.8%
  • C 1.5%