Riedones3D
It is the code to perform coin die recognition and die clustering
Installation
requirements:
pyrender==0.1.45
hydra-core==1.0.0
omegaconf==2.0.6
scikit-learn==0.24.2
torch-cluster==1.5.9
torch-scatter==2.0.7
torch-geometric==1.7.2
torch==1.8.1
torch_points_kernels
torch-points3d==1.3.0
MinkowskiEngine==0.5.2
torch-sparse==1.4.0
omegaconf==2.0.6
hydra-core==1.0.0
open3d==0.12.0
First install anaconda or miniconda (use this website for example)
then open a terminal.
create a new environnement (for example using conda)
conda create -n "riedones_3d"
conda activate riedones_3d
execute this script to install the correct packages(here we install with cuda102)
pip install pyrender
pip install trimesh
pip install torch==1.8.1
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-cluster -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-spline-conv -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-geometric==1.7.2
pip install git+https://github.com/nicolas-chaulet/[email protected]
install minkowski engine
apt-get install build-essential python3-dev libopenblas-dev
pip install -U git+https://github.com/NVIDIA/Mink[email protected] --no-deps
install torch-sparse
apt-get install libsparsehash-dev
pip install --upgrade git+https://github.com/mit-han-lab/[email protected]
Preprocessing
As input, we need point cloud in ply format. We propose a simple script to convert from stl to ply with normals(if you already have point cloud in ply, you do not need this step).
python scripts/mesh2pcd.py --path_coin mymesh.stl
Register a pair of coin
It will register the pair and it will also compute the histogram of distance. It displays the results with open3D.
python scripts/whole_pipeline.py --path COIN1.ply COIN2.ply -m PATH OF THE MODEL --angle 0 --trans 20 --clf classifiers/logistic_part_droits_sym.pkl --path_scaler classifiers/mean_std.json --est ransac
Coin die Clustering
Compute the features
First you need to compute the features:
python scripts/compute_feature.py --path_coin DROITS --list_coin Coins_et_Monnaies_Droits_all.csv -m PATH OF THE MODEL --path_output results --name Droits
It takes few minutes to compute every features
compute a pair similarity comparison
python scripts/compute_transformation.py --path_feature results/Droits/feature/ --path_output results/Droits/transformation --list_coin Coins_et_Monnaies_Droits_all.csv --num_points 5000 --est ransac --clf classifiers/logistic_part_droits_sym.pkl --path_scaler classifiers/mean_std.json --n_jobs 8 --sym
--sym means the histogram is symmetric. --num_points is the number of points --est is the robust estimator to compute the transformation It takes few days to compute every similarities it will generate two files:
- a file containing every transformations
- a file containing every histograms of distance
Compute the Graph
We compute the graph of similarity between the pairs of coins
python scripts/compute_graph_from_hist.py --path_histogram results/Droits/transformation/hist.npy -m classifiers/logistic_part_droits_sym.pkl -o results/Droits/graph --path_scaler classifiers/mean_std.json
Clean the graph using graph visualizer
You can select links nodes, remove/add links, search for a node cluster the results, save the graph.
TODO: Tutorial about how to use graph visualizer
generate images and 3D models
TODO: Tutorial for this script
1 Jun 15, 2022
3 Nov 28, 2021
0 Feb 07, 2022
1 Jan 25, 2022
2 Oct 16, 2022
2 Dec 16, 2021
48 Jan 28, 2021
74 Sep 13, 2022
194 Jan 02, 2023
4 Oct 18, 2021
4 Oct 31, 2022
631 Jan 05, 2023
23 Dec 29, 2022
1 Apr 18, 2022
0 Nov 27, 2021
7 Dec 29, 2022
7 Nov 12, 2022
59 Dec 19, 2022
1 Jan 03, 2023
2 Jan 17, 2022