---

pyDBoW3

Ultra-fast Boost.Python interface for DBoW3

This repo was created in order to interface DBoW algorithm from python in another project EasyVision. It is being used for a simple topological SLAM implementation since OpenCV BowKMeansTrainer doesn't work with binary features.

If you wish you use it on your own it is as easy as:

import pyDBoW3 as bow
voc = bow.Vocabulary()
voc.load("/slamdoom/libs/orbslam2/Vocabulary/ORBvoc.txt")
db = bow.Database()
db.setVocabulary(voc)
del voc
# extract features using OpenCV
...
# add features to database
for features in features_list:
  db.add(features)

# query features
feature_to_query = 1
results = db.query(features_list[feature_to_query])

del db

This repository was created based on pyORBSLAM2 and ndarray to cv::Mat conversion on numpy-opencv-converter.

Note

Tested on these platforms:

• OpenCV 3.4.2.16
• Windows 10 msvc 2017 x64
• xenial with Python 2.7, libboost 1.54 (autobuild with travis)
• xenial with Python 3.5, libboost 1.54 (autobuild with travis)

Get started

Windows

Prerequisites: * OpenCV * Python with Numpy and opencv-contrib-python * Boost >1.54 * cmake * Microsoft Visual Studio

To build Boost.Python, go to Boost root and run:

bootstrap.bat --prefix=/dir/to/Boost.Build

Then build Boost.Python like this:

/dir/to/Boost.Build/b2 --with-python threading=multi variant=release link=static

To build DBoW3, simply run build.bat file and then build solution folder in install/DBoW3/build and then the solution in build folder.

Currently there is no python package generation, so you could simply copy pyDBoW3.pyd and opencv_world*.dll files to your virtual environment.

Unix

Use build.sh to build build/pyDBoW.so, which you should then put on your PYTHONPATH.

Check .travis.yml for environment variables.

Mac OSX

Use build.sh to build build/pyDBoW.so, which you should then put on your PYTHONPATH.

Check .travis.yml for environment variables.

Note

You will probably need to run sudo make install for install/opencv/build to install it on your system.

Using under a conda environment (to use pre-installed version of OpenCV)

Build a conda environment, adding boost and cmake to it, something like:

conda create -n test_env python=3.5 opencv=3.3.1 cmake boost matplotlib numpy

## Unlink system boost installed by brew
brew unlink boost

Use build_under_conda.sh to build build/pyDBoW.dylink which is symlinked to build/pyDBoW.so. Add this to your conda environment by creating a .PTH file under /Users/<user_name>/anaconda3/envs/test_env/lib/python3.5/site-packages/pydbow3.pth containing /Users/<user_name>/pyDBoW3/build