Intro

Doppia is a collection of several research publications and original repo is hosted on bitbucket.

Docker Run

Well, if you are too lazy to read this whole readme, going through all the deatils of setting up the machine, compile the C++ applications and run demo with various flags, fine, I made a docker version for you to just execute 2 lines of code doing everything for you:

# First of all, git clone this repo and in here we assume, you navigated to the root of this repo
# Then, just simply build the docker image by running

make build

# Then, it will take a while and build a docker image at around 8 GB, thanks for opencv, then you just simply run it

make run

And then, it will spin up a container with ground_estimation app compiled. You can run the demo by:

./ground_estimation -c test.config.ini --gui.disable true

Then you are supposed to see logs below:

Ground estimation. Rodrigo Benenson @ KULeuven. 2010-2011.
Parsed the configuration file test.config.ini
Creating new log file stixel_world.out.txt
Using stereo camera calibration file: ../../video_input/calibration/stereo_calibration_bahnhof.proto.txt
2017-09-11 08:46:07 {7f388f16f800} [ StereoCameraCalibration ] : stereo_calibration_data name: Andreas Ess Bahnhofstrasse sequence stereo calibration
Reading files:
../../../data/sample_test_images/bahnhof/image_00000000_0.png
../../../data/sample_test_images/bahnhof/image_00000000_1.png
Average iteration speed  39.3800 [Hz]  (in the last 10 iterations)
Requested frame number 11 but frames should be in range (0, 10)
Processed a total of 10 input frames
End of game, have a nice day.

It means you are successfully running the demo.

If you wanna do more, then continue within this docker container and I strongly suggest you also read rest of the readme to get a better idea.

Machine Setup

• Ubuntu (preferred)

• GPU hardware

• CUDA with capacity to be higher than 2.0

• All boost libraries

• Google protocol buffer

• OpenCV 2.4 (3.0 is not supported yet)

• libjpeg, libpng

• libSDL

• CMAKE >= 2.4.3

Dependencies	Version	Installation
CUDA	8.0	Follow this procedures
OpenCV	2.4.13	Run this script
Protobuf	latest	Run this script
All boost libraries	latest	sudo apt-get install libboost-all-dev
libjpeg	latest	sudo apt-get install libjpeg-dev
libpng	latest	sudo apt-get install libpng-dev
libSDL	latest	sudo apt-get install libSDL-dev

Compile and Run Demo

The code includes many applications and they all locate under src/applications.

Before doing everything, you need to modify this line in common_settings.cmake to replace the placeholder with your machine name.

Then run following script to ensure the protocol buffer files match the version installed in your system.

generate_protocol_buffer_files.sh

0. Summary

Basically, this is a project based on C++, meaning, how it works is that you need to 1) compile the code and obtain an executable, 2) config parameters and run executable against input.

Therefore, we describe compile (ground_estimation/stixel_world as CPU only example and objects_detection as GPU example) first then how to run the demo.

1. Compile

1.1 CPU Only Compile

Example for ground_estimation app

# Navigate to the path
cd doppia/src/applications/ground_estimation

# Compile
cmake -D CMAKE_BUILD_TYPE=RelWithDebInfo . && make -j2

Example for stixel_world

cd doppia/src/applications/stixel_world
cmake . && make -j2 

1.2 GPU Compile

objects_detection is build in a way to work with GPU

cd doppia/src/applications/objects_detection
cmake -D CMAKE_BUILD_TYPE=RelWithDebInfo -D USE_GPU=ON . && make -j2

2. Run Demo

2.1 Run Ground Estimation Demos

./ground_estimation -c test.config.ini

2.2 Run Stixel World Demos

# Demo 1
OMP_NUM_THREADS=4 ./stixel_world -c fast.config.ini --gui.disable false

# Demo 2
OMP_NUM_THREADS=4 ./stixel_world -c fast_uv.config.ini --gui.disable false

2.3 Run Object Detection Demos

# Demo 1
OMP_NUM_THREADS=4 ./objects_detection -c cvpr2012_very_fast_over_bahnhof.config.ini --gui.disable false

# Demo 2
OMP_NUM_THREADS=4 ./objects_detection -c cvpr2012_inria_pedestrians.config.ini --gui.disable false

# Demo 3
OMP_NUM_THREADS=4 ./objects_detection -c cvpr2012_chnftrs_over_bahnhof.config.ini --gui.disable false

# Demo 4
./objects_detection -c eccv2014_face_detection_pascal.config.ini --gui.disable false

2.4 Summary

So basically, after compile an app, you just need to define a config file with format like *.config.ini and use objects_detection executable to run against it.

For example, pedestrians detection has an example config file and in this file, you can define whether you wanna input from directory or camera, scales, stride, mask, ...etc. For the example config of pedestrians detection, the input is sourced from directory, which you can also easily define your own ones.

For more details, you can run ./objects_detection --help to see all the options.

For example, if you want to save the detections and save screenshots into local folder then you can use --save_detections and --gui.save_all_screenshots flags:

./objects_detection -c ***.ini --save_detection=1 --gui.disable=false --gui.save_all_screenshots=1

You can find the detailed info by running ./objects_detection --help

3. Modifications

3.1 OpenCV issues

Due to some OpenCV linking issues originally potentially from wrong order of linking lib, we have a fix for compiling apps by commenting out following in all CMakeLists.txt. (Original issue discussion can be found here).

#set(opencv_LIBRARIES
#    opencv_core opencv_imgproc opencv_highgui opencv_ml
#    opencv_video opencv_features2d
#    opencv_calib3d
#    opencv_objdetect opencv_contrib
#    opencv_legacy opencv_flann
#    opencv_gpu
#   ) # quick hack for opencv2.4 support

Example changes:

Name	Line
ground_estimation	line_40
stixel_world	line_40
objects_detection	line_42

3.2 CUDA modifications

You need to modify the cuda path with correct version on your machine and in here, the changes are below and original discussions can be found here:

• Replacing all cuda-5.5 with cuda-8.0 since that's what I tested (you may have a different version and check by ls /usr/local | grep cuda)

• Replacing all cuda-5.5 in all related CMakeList.txt, for example in objects_detection

• Adding USE_GPU flag in objects_detection CMakeList.txt

• Indexing CUDA arch specifically in objects_detection CMakeList.txt and make sure that the arch is the one your host machine is compatible, otherwise you may still be able to compile but throw errors

• Enforce objects_detection to be build with GPU ON by adding flag in cmake command