pandora 1.2.1

Pandora is a stereo matching framework that helps emulate state of the art algorithms

A stereo matching framework that will help you design your stereo matching pipeline with state of the art performances.

Overview • Install • First Step • Customize • Credits • Related • References

Overview

From stereo rectified images to disparity map	Pandora is working with cost volumes

Pandora aims at shortening the path between a stereo-matching prototype and its industrialized version.
By providing a modular pipeline inspired from the (Scharstein et al., 2002) taxonomy, it allows one to emulate, analyse and hopefully improve state of the art stereo algorithms with a few lines of code.

We (CNES) have actually been using Pandora to create the stereo matching pipeline for the CNES & Airbus off board processing chain.
Leaning on Pandora's versatility and a fast-paced constantly evolving field we are still calling this framework a work in progress !

Install

Pandora is available on Pypi and can be installed by:

pip install pandora

For stereo reconstruction we invite you to install pandora and the required plugins using instead the following shortcut:

pip install pandora[sgm, mccnn]

First step

Pandora requires a config.json to declare the pipeline and the stereo pair of images to process. Download our data sample to start right away !

• cones stereo pair
• a configuration file

# install pandora latest release
pip install pandora

# download data samples
wget https://raw.githubusercontent.com/CNES/Pandora/master/data_samples/images/cones.zip  # input stereo pair
wget https://raw.githubusercontent.com/CNES/Pandora/master/data_samples/json_conf_files/a_local_block_matching.json # configuration file

# uncompress data
unzip cones.zip

# run pandora
pandora a_local_block_matching.json output_dir

#Left (respectively right) disparity map is saved in output_dir/left_disparity.tif (respectively output_dir/right_disparity.tif)

To go further

To create you own stereo matching pipeline and choose among the variety of algorithms we provide, please consult our online documentation.

You will learn:

• which stereo matching steps you can use and combine
• how to quickly set up a Pandora pipeline
• how to add your own private algorithms to customize your Pandora Framework
• how to use Pandora API (see CARS for real life exemple)

Credits

Our data test sample is based on the 2003 Middleburry dataset (D. Scharstein & R. Szeliski, 2003).

(D. Scharstein & R. Szeliski, 2002). Scharstein, D., & Szeliski, R. (2002). A taxonomy and evaluation of dense two-frame stereo correspondence algorithms. International journal of computer vision, 47(1-3), 7-42.
(D. Scharstein & R. Szeliski, 2003). Scharstein, D., & Szeliski, R. (2003, June). High-accuracy stereo depth maps using structured light. In 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003. Proceedings. (Vol. 1, pp. I-I). IEEE.

Related

Plugin_LibSGM - Stereo Matching Algorithm plugin for Pandora
Plugin_MC-CNN - MC-CNN Neural Network plugin for Pandora
CARS - CNES 3D reconstruction software

References

Please cite the following paper when using Pandora:
Cournet, M., Sarrazin, E., Dumas, L., Michel, J., Guinet, J., Youssefi, D., Defonte, V., Fardet, Q., 2020. Ground-truth generation and disparity estimation for optical satellite imagery. ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences.