OneACPose - Relative Pose from Deep Learned Depth and a Single Affine Correspondence

Code for our ECCV 2020 paper: Relative Pose from Deep Learned Depth and a Single Affine Correspondence Ivan Eichhardt, Daniel Barath; The European Conference on Computer Vision (ECCV), 2020

Access the paper and supplementary material at ecva.net.

Cite it as

@InProceedings{Eichhardt_Barath_2020_ECCV,
	author = {Eichhardt, Ivan and Barath, Daniel},
	title = {Relative Pose from Deep Learned Depth and a Single Affine Correspondence},
	booktitle = {The European Conference on Computer Vision (ECCV)},
	month = {August},
	year = {2020},
	doi = {10.1007/978-3-030-58610-2_37}
}

Build using CMAKE

See BUILD for dependencies and instructions to build our C++ library and sample projects.

Python bindings

To install Python bindings (package called "pyoneacpose") use the following commands:

$ git clone --recursive https://github.com/eivan/one-ac-pose.git
$ pip install ./one-ac-pose

To use the package, launch python and import "pyoneacpose"

$ python

>>> import pyoneacpose

Usage

For python:

• Install Python bindings (see above)
• Open Jupyter Notebook from the root of the cloned repository
• Launch either of the example scripts

Example Jupyter Notebooks:

• Piccadilly scene from the 1DSfM dataset
• Fountain scene from the Strecha dataset