This repository contains the comple code of an end-to-end pipeline used to generate the graph-based scene description. A single, moblie RGBD-camera is used to capture the scene. The software runs on a SCITOS G5 from MetraLabs at either 10Hz or 30Hz. The robot is equipped with an eight core Intel Core i7-6700T CPU running at 2.8GHz and 8GHz RAM.
- RGBD-camera (Code is based on libfreenect2 and uses a Microsoft Kinect v2)
Software needed to build:
- non-comercial MIRA-framework
- python (required to run evaluation scripts)
Software downloaded by the cmake script:
- OpenCV 4.3.0 + contrib
- bazel 2.0.0
- tensorflow r2.2
- protobuf 3.8.0
- neo4j (server executable, java required)
- libneo4j-client
- PCL
- libfreenect2
- rapidjson (for evaluation only)
Neural network models:
- R-CNN model is provided
- YOLO (optional) download v3 .cfg-file here and .weights here, put files into
external/models/YOLO
Run
git clone https://github.com/Ne94fets/SceneGraphMapGen.git
mkdir build
cd build
cmake ../
make
Instead of running cmake you can also import the CMakeLists.txt in your favourite Editor, like qtcreator.
see here
see here