The objective of our project is to enable an efficient search in a 3D model bank models from simple hand drawings. The project is based on this paper SIGGRAPH2012.
The project lies on differents techniques such Suggestive Contouring (Canny filter), Gabor filtering (gaussian convolution on Fourier Transformation of the input image) and histogram representation using Visual Vocabulary.
Below is the complete search pipeline for matching meshes :
The external C++ libraries needed are:
- OpenCV
- Boost
- OpenGL
- PythonLibs
You also need to download the igl libray. Finally your active Python environment needs VTK, numpy, sklearn, and opencv.
clone the repository using git, and build it with CMake
git clone https://github.com/PierreTsr/Shape_Retrieval.git
cd Shape_Retrieval/
mkdir build
cd build/
cmake ../
make
The project needs to be cloned next to the igl library to work. The Princeton Shapes dataset must be pasted next to the project too.
The first step is to compute the 2D renders of the model using
python ../OccludingContours/generateDataset.py
Then running ./ex_bin will try to generate the histograms of a part of the dataset, load them, and display a window to draw in. The closest views' ID will be returned in the console.
Many path are hard-coded in pre-compilator variables and need to be respected for the program to work.