Requirements: Python 3 Dependencies: numpy, keras, keras-retinanet, PIL, matplotlib, easygui, opencv-python, csv, seaborn, tensorflow
Anaconda distributive recommended since it contains most of packages preinstalled packages can be installed via command
pip install packagename
windows executable produced by pyinstaller can be downloaded here http://friendgame.byethost24.com/js/SEM_analysis/SEM_analysis.zip
Colab version can be used from here
https://colab.research.google.com/drive/1wQAnWgIyhI-VEGI0y22R1cuPIhNGmMVH?usp=sharing
SEM_analysis neural network Download inference from google drive https://drive.google.com/file/d/1qduzSqZ6V7J-qpSX5C6Ly1OIwW417kt6/view?usp=sharing Place it to the same folder as inference.py script and run
python inference.py
For own dataset synthesis:
- place files with your to textures folder
- run jupyter_NPS_generate.ipynb for JSON files generation (generated JSON contains particles positions and sizes)
- run blender_powder.py into blender enviroment
- It will generate annotations.csv file with all labels and images at directory "renders"
Detailed RetinaNet trainig procedure can be found at https://www.kaggle.com/alexanderkhar/nps-detector Example of generated dataset can be found at https://www.kaggle.com/alexanderkhar/generated-nps