monodepth-cpp

Tensorflow C++ implementation for single image depth estimation

The original work is implemented in Python, click here to go to their repo. Please cite their work if your find it helpful.

It is using pointer-to-implementation technique, so that you can use it in your project without worrying the actual implementation. Refer to src/inference_monodepth.cpp for more information

The C++ version is about 28fps, in comparison with Python's 13fps, tested with i7 processor and NVidia 1070 graphics laptop.

Related repo

If you are looking for single-image relative depth prediction, feel free to check out my other repo, MiDaS-cpp, a PyTorch C++ implementation of MiDaS.

Requirements

Option 1

Run the Docker container

chmod +x run_tf_docker.sh
./run_tf_docker.sh

Note: You may have a better graphics card, and if that's so, go to tf.Dockerfile line 55 and change the value of TF_CUDA_COMPUTE_CAPABILITIES (check out Nvidia cuda-gpus ).

Option 2

Build Tensorflow library from source, see here

Freeze TF graph

Either way, you will need to Freeze a Tensorflow graph (with known input and output names), or download a converted graph here.

Configure CMake project

• Make sure you have the compatible versions of Eigen, Protobuf, and Tensorflow (Mine: Eigen 3.3.4; Protobuf 2.6.1-1.3; Tensorflow 1.6)
• If you have a set of different paths to the libraries, modify the .cmake files in CMakeModules

Build (static/shared) library

NOTE: If you use the Docker container, this github repo will be mounted in /monodepth. Therefore, run cd monodepth in the container first.

mkdir build && mkdir install
cd build
cmake -DCMAKE_INSTALL_PREFIX=../install ..
make && make install

You will be seeing 'include' and 'lib' folders in the 'install' folder, import them in your project

To test if Monodepth C++ is working properly,

cd build
./inference_monodepth

NOTE: Select either static or shared library in CMakeLists.txt, unless you want both of them

Use (static/shared) library

In your own CMakeLists.txt project, do the following:

set(monodepth_INCLUDE_DIRS /path/to/monodepth-cpp/install/include)
INCLUDE_DIRECTORIES(
  ...
  monodepth_INCLUDE_DIRS
  ...
)

TARGET_LINK_LIBRARIES(awesome_exe /path/to/tensorflow/library/libtensorflow_cc.so) # Only if you are using the provided instructions
TARGET_LINK_LIBRARIES(awesome_exe /path/to/monodepth-cpp/install/lib/libmonodepth_static.a) # if you are using static library

Personal projects that use monodepth-cpp

1. CNN-SVO

2. Reproducing Deep virtual stereo odometry

NOTE: Because DVSO uses both the left disparity and the right disparity outputs (for left-right consistency check), it requires some modifications in the source code to enable the disparities outputs.