AtomAI: Reusability

Original AtomAI repository (up to date)

Original AtomAI repository (fork used)

Files contributed

The following files are the main files contributed by this work, the path of the files are provided below. (init and other small changes are not included in this list)

• atomai/utils/instance_seg_utils.py: "Introduced utilities for Instance Segmentation"

• atomai/transforms/imaug.py: "Fixed image augmentation to access all flips and rotations"

• atomai/trainers/trainer.py: "Introduced Early Stopping and LSTM trainer for ImSpec"

• atomai/nets/ed.py: "Signal LSTM introduced"

• atomai/models/segmentor.py: "Introduced Binary Automatic-Thresholding and changes for Early Stopping "

• atomai/models/loaders.py: "Changes made for Binary Automatic-Thresholding"

• atomai/models/imspec.py: "ImSpecTrainerLSTM introduced"

• Mat2Spec/

  • Mat2Spec/Mat2Spec.py: "CNN and VAE modules for spectral data analysis"
  • Mat2Spec/SinkhornDistance.py: "Sinkhorn distance implementation"
  • Mat2Spec/data.py: "Data loading with encoding and clustering"
  • Mat2Spec/file_setter.py: "Dataset preparation based on input property and source"
  • Mat2Spec/pytorch_stats_loss.py: "Implementation of differentiable statistical distance losses"
  • Mat2Spec/utils.py: "Utility functions for training, evaluation, and model handling"

• examples/

  • examples/notebooks/EarlyStopping_HRTEM_pv.ipynb
  • examples/notebooks/EarlyStopping_LBFO_pv.ipynb
  • examples/notebooks/Instance_Segmentation.ipynb
  • examples/notebooks/LSTM_EarlyStopping_ImSpec.ipynb
  • examples/notebooks/Mat2Spec_ImSpec.ipynb

To reproduce our results, please refer to the following Jupyter notebooks:

Example 1: LBFO dataset demo

Implementation of Updated Deep Convolutional Neural Network for Atom Finding using Early stopping

Example 2: HRTEM dataset (Au/CdSe/Combined) demo

Implementation of Updated DCNN equipped with Early stopping and automatic binary-thresholding for Nano-particle Segmentation

Example 3: Instance Segmentation: Additive Manufacturing dataset demo

Implementation of Semantic Segmentation from Semantic Segmentation maps for optical micrographs of Additive manufacturing dataset

Example 4: ImSpec demo

Implementation of Updated Demo for ImSpec on the STEM EELS dataset using Early Stopping and LSTM Decoder

Example 5: Mat2Spec demo

Implementation of the modified Mat2Spec for ImSpec on the STEM EELS dataset

The following configurations were used for training

The parameters like training cycles, patience, and the number of augmented images created are modified so as to run in a Google Colab notebook without any premium resources. The training cycles were set to 5000 and patience to 500, and the number of images produced was ~ 2500.

One only needs to run the Colab notebooks as is. However, if one wants to test it on a personal machine, they can download the necessary datafiles for Example 1 and 2 here.

The additive manufacturing dataset can be obtained here.

The STEM Images can be downloaded here

Credits and rights for the dataset used in each file belong to the respective entities:

• LBFO: C. Nelson, A. Ghosh, M. Ziatdinov and S. Kalinin V, (Zenodo, 2021).
• Au &CdSe: Groschner, C., Choi, C., & Scott, M. (2021). Machine Learning Pipeline for Segmentation and Defect Identification from High-Resolution Transmission Electron Microscopy Data. Microscopy and Microanalysis, 27(3), 549-556. doi:10.1017/S1431927621000386
• Additive Manufacturing Dataset: Can be provided upon request to the authors.
• STEM-EELS: Roccapriore, Kevin M. and Ziatdinov, Maxim and Cho, Shin Hum and Hachtel, Jordan A. and Kalinin, Sergei V. (2021). Predictability of Localized Plasmonic Responses in Nanoparticle Assemblies. doi:10.1002/smll.202100181