FastGMC 🚀

This repo provides an implementation of global camera compensation (GMC) using OpenCVs Video Stabilization classes. As these are not available for Python directly, we provide a C++ Wrapper using modern and seamless Numpy integration through Pybind11.

Usage

import numpy as np
import cv2 as cv
from fast_gmc import gmc

prev = cv.imread('warp_test_0.jpg')
curr = cv.imread('warp_test_1.jpg')

# Run the actual gmc
mat = gmc(curr, prev, downscale=4, model='affine')  # or model='homography'

print("Transformation matrix:\n{}".format(mat))

How to install

Installation can be done through pip or by building the extensions directly.

Dependencies

This build depends CMake (>=3.0).

Local install (Linux / Windows)

Download and install from CMake website: official site Be sure to update your environment variables to contain your downloaded CMake. On Linux this can be done with export PATH=<path-to-cmake-dir>/bin:$PATH.

Global install (Linux)

sudo apt-get install cmake CMake will be automatically added to PATH variable.

Install using PIP (recommended)

This works currently ONLY inside a Conda environment.

pip install git+https://github.com/DLR-MI/fast_gmc.git

Install using CMake (Advanced)

git clone https://github.com/DLR-MI/fast_gmc.git gmc_pybind
cd gmc_pybind/src/fast_gmc/build/
cmake ../
cmake --build .
# You can now copy the *.so file and use it in your project.

Testing

You can test if the installation worked by doing the following:

git clone https://github.com/DLR-MI/fast_gmc.git gmc_pybind
# run a test (optional)
pip install opencv-python
python gmc_pybind/test/test.py
# Expected output:
# Previous to current frame = 12.25531005859375 (RMSE)
# Warped to current frame = 9.708952903747559 (RMSE)
# Transformation matrix:
# [[ 1.0105095e+00 -3.5821529e-05 -1.0788696e+01]
#  [ 3.5821529e-05  1.0105095e+00 -6.7262573e+00]
#  [ 0.0000000e+00  0.0000000e+00  1.0000000e+00]]