This reflects the following work: D. Alina, A. Shomanov and S. Baimukhametova, "MaLeFiSenta: Machine Learning for FilamentS Identification and orientation in the ISM," in IEEE Access, 2022, doi: 10.1109/ACCESS.2022.3189646.
Contains
Details: This is not a fully automatized code. One should define data directories and build the training datasets. This code then prepares the datasets for the neural network models. The models set-up are also presented : Mask RCNN and Unet. The optimization procedures are also included. Moreover, the Morphologcal Similarity Index Measurement functions (Wang et al.2005, Green et al.2017) are also included which allows a user to make comparison, if necessary. The models saving is also included.
Update (Jul 29, 2023): due to the inconsistencies between the recent distributions of tensorflow and maskrcnn, we recommend the following solution: see https://github.com/KenessaryMurat/Mask-R-CNN-for-Filaments-Identification.git
Update (Aug 17, 2024): the latest version, with the compiled jupyter notebook and model weights is located here: https://github.com/Temirkul/ism-filaments-analysis