Skip to content

Latest commit

 

History

History
110 lines (81 loc) · 3.92 KB

README.md

File metadata and controls

110 lines (81 loc) · 3.92 KB

Python Bundle Adjustment

These instructions are based on and almost identical to the ones at https://github.com/drormoran/gasfm/tree/main/bundle_adjustment/README.md.

Conda envorinment

Use the gasfm environment.

conda activate deepaat
export PYBIND11_PYTHON_VERSION="3.8"
export PYTHON_VERSION="3.8"

Directory structure

After this set up, the directory structure be:

DeepAAT
├── bundle_adjustment
│   ├── ceres-solver
│   │   ├── ceres-bin
│   │   |   └── lib
│   │   |       └── PyCeres.cpython-39-x86_64-linux-gnu.so
│   │   ├── ceres_python_bindings
│   │   |   └── python_bindings
│   │   |       └── custom_cpp_cost_functions.cpp
│   │   └── CMakeLists.txt
│   └── custom_cpp_cost_functions.cpp
├── code
├── datasets

Set up

  1. Clone the Ceres-Solver repository to the bundle_adjustment folder and check out version 2.1.0:
cd bundle_adjustment
git clone https://ceres-solver.googlesource.com/ceres-solver -b 2.1.0
  1. Clone the ceres_python_bindings package inside the ceres-solver folder:
cd ceres-solver
git clone https://github.com/Edwinem/ceres_python_bindings
  1. Copy the file "custom_cpp_cost_functions.cpp" and replace the file "ceres-solver/ceres_python_bindings/python_bindings/custom_cpp_cost_functions.cpp". This file contains projective and euclidean custom bundle adjustment functions.
cp ../custom_cpp_cost_functions.cpp ceres_python_bindings/python_bindings/custom_cpp_cost_functions.cpp

Next, you need to build ceres_python_bindings and ceres-solver and create a shared object file that python can call. You can either continue with the instructions here or follow the instructions at the ceres_python_bindings repository.

  1. run:
cd ceres_python_bindings
git submodule init
git submodule update
  1. Make sure that the C++ standard library version used during the build is recent enough, and not hard-coded to C++11 by pybind11. Please check your c++ compiler version and modify it bellow (for example here is c++17):
sed -i 's/set(PYBIND11_CPP_STANDARD -std=c++11)/set(PYBIND11_CPP_STANDARD -std=c++17)/g' AddToCeres.cmake
  1. Add to the end of the file ceres-solver/CMakeLists.txt the line: "include(ceres_python_bindings/AddToCeres.cmake)":
cd ..
echo "include(ceres_python_bindings/AddToCeres.cmake)" >> CMakeLists.txt
  1. Inside ceres-solver folder run:
mkdir ceres-bin
cd ceres-bin
cmake ..
make -j8
make test
  1. If everything worked you should see the following file:
bundle_adjustment/ceres-solver/ceres-bin/lib/PyCeres.cpython-39-x86_64-linux-gnu.so
  1. If you want to use this bundle adjustment implementation for a different project make sure to add the path of the shared object to linux PATH (in the code this is done for you). In the python project this would be for example:
import sys
sys.path.append('../bundle_adjustment/ceres-solver/ceres-bin/lib/')
import PyCeres

To see the usage of the PyCeres functions go to code/utils/ceres_utils and code/utils/ba_functions.

Note

If you encounter problems while compiling PyCeres, you can also refer to GASFM's bundle adjustment instruction or create a new environment for compilation. If the environment for compiling PyCeres is not the same as the network environment, then run_ba in conf file can be set to False during training and inference, and after obtaining the predicted results, run run_ba.py separately for BA.