-------------------------------------------------------README file for------------------------------------------------------ Python code: Classification of in situ high speed videos of the gravure printing fluid splitting process using deep learning ---------------------------------------------------------------------------------------------------------------------------- Authors of TUdatalib submission: Pauline Brumm, Nicola Ciotta, Hans Martin Sauer, Andreas Blaeser, Edgar Doersam Technical University of Darmstadt, Department of Mechanical Engineering, Institute of Printing Science and Technology (IDD) and BioMedical Printing Technology (BMT), Magdalenenstr. 2, 64289 Darmstadt, Germany Last modified: 2022-08-26 (yyyy-mm-dd) ----------------------------------- 1. GENERAL In this TUdatalib submission, you will find the Python code and additional data as used in the following publication: Pauline Brumm, Nicola Ciotta, Hans Martin Sauer, Andreas Blaeser and Edgar Doersam Deep learning study of induced stochastic pattern formation in the gravure printing fluid splitting process Journal of Coatings Technology and Research, accepted for publication in July 2022 https://doi.org/10.1007/s11998-022-00687-x Please refer to the publication for further information. 2. MANUAL FOR USING THE PYTHON CODE - Download and unzip folder "Python_code". The required Python packages and manual for installation can be found in the file "requirements.txt". - Use "requirements-orig.txt" instead of "requirements.txt" if you want to use the exact same python package versions as used in our publication. However, we recommend to use "requirements.txt" as you get the newest packages from that and as it can be easily installed. - Download and unzip folder "Frame_dataset_PNG_8_bit" and copy it into the folder "Python_code" to be able to train your own deep learning models on the Schaefer dataset using the "training.py" script. - Download and unzip folder "models" and copy it into the folder "Python_code" to be able to use several already trained models for the script "inference.py". 3. WHAT DOES THE PYTHON CODE DO? The code consists of several Python scripts, which are described in the following: create_frame_dataset_png.py --> Extracts a frame dataset from the Schaefer dataset. The result of this operation is provided as the folder "Frame_dataset_PNG_8_bit". pyMRAW_modified.py --> Contains a helper function for "create_frame_dataset_png.py". training.py --> Trains deep learning models on extracted frames from the Schaefer dataset. The trained models are able to classify the videos from the Schaefer dataset into three classes of pattern formation phenomena. inference.py --> Uses the trained deep learning models for inference. nicola_utils.py --> Contains helper functions for "training.py" and "inference.py". pytorch_cookbook_utils_modified.py --> Contains helper functions for "training.py" and "inference.py". Besides, the file "Dataset_labels.csv" is provided in the folder "Python_code". It contains the labels for all videos as defined in the Schaefer dataset. The Schaefer dataset can be found at: https://doi.org/10.25534/tudatalib-191.2 Julian Schaefer, 2019: Messdaten und Auswertungssoftware fuer die hochdynamische Grenzflaecheninstabilitaet im Zylinderspalt. TUdatalib. 4. ADDITIONAL FILES TO DOWNLOAD In addition to the Python code, the frame dataset and the trained models, you can download the following files: results.zip --> Contains some training and inference details for all models. maps.zip --> Contains exemplary class activation maps for each layer (l1, l2, l3, l4) of one exemplary model. Refer to publication for more details. class_probabilities.zip --> Contains boxplots of class probabilities for several models and Python code to create them. Refer to publication for more details. 5. THE AUTHORS Pauline Brumm: Supervised bachelor thesis of Nicola Ciotta and optimized data and code for upload to TUdatalib; PhD candidate at IDD Nicola Ciotta: Developed the Python code in his bachelor thesis "Development of a computer vision workflow for high speed videos" in 2020/2021 at IDD Hans Martin Sauer: Supervisor of the research project; head of the research group Printing Fluids and Interfaces at IDD Andreas Blaeser: Head of BMT Edgar Doersam: Head of IDD and PI of the research project 6. ACKNOWLEDGMENTS We kindly acknowledge the financial support by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - Project-ID 265191195 - Collaborative Research Center 1194 (CRC 1194) "Interaction between Transport and Wetting Processes", project C01. 7. LICENSE This data collection is released under the creative commons license condition of CC BY-NC.