How leticiarosimporto/msi-deeplearning works
MSI-DeepLearning processes multispectral imagery from band registration and HDF5 dataset creation through model training and evaluation, with spectral inspection as an analysis branch. The README documents the overall workflow; excerpts show registration utilities, HDF5 loading, augmentation, model definitions, and training. The dataset-builder and several analysis/training entry points were not excerpted, so their connections are limited to documented relationships.
Main entry points: training.py Training loop · register_MSI_ECC.py ECC registration · msi_dataset.py MSI dataset loader · parrallel_ECC.py Parallel ECC
Components 18
Band registration
- ECC registration
register_MSI_ECC.py - Parallel ECC
parrallel_ECC.py
Dataset preparation
- Image utilities
ms_utils/ms_tools.py - HDF5 builder
make_msi_dataset.py - Cube serialization
ms_utils/msi_image_utils.py - HDF5 cubes
- MSI dataset loader
deep_utils/msi_dataset.py - Hyperspectral loader
deep_utils/HsiDataset.py - Band-preserving augmentation
deep_utils/data_augment.py - Cube utilities
ms_utils/msi_image_utils.py
Model learning
- ResNet models
deep_utils/ResNets.py - Depthwise ResNet
deep_utils/MSI_RESNET.py - Training loop
deep_utils/training.py - Model evaluation
deep_utils/testing.py
Spectral analysis
- Spectral extraction
h5_to_spectral_mean.py - Spectral tools
ms_utils/ms_tools.py
Other components
- Researcher
- Raw band images
Connections 18
- Researcher → ECC registration: registers bands
README.md - Researcher → Parallel ECC: registers in parallel
README.md - Raw band images → ECC registration: supplies images
- ECC registration → Image utilities: assembles bands
register_MSI_ECC.py - Parallel ECC → Image utilities: assembles bands
parrallel_ECC.py - Researcher → HDF5 builder: builds dataset
README.md - HDF5 builder → Cube serialization: saves datacubes
- Cube serialization → HDF5 cubes: writes cubes
ms_utils/msi_image_utils.py - MSI dataset loader → HDF5 cubes: reads datacubes
deep_utils/msi_dataset.py - Hyperspectral loader → HDF5 cubes: reads samples
deep_utils/HsiDataset.py - MSI dataset loader → Band-preserving augmentation: applies augmentation
- Researcher → Training loop: trains model
README.md - Training loop → ResNet models: updates parameters
deep_utils/training.py - Training loop → Depthwise ResNet: updates parameters
- Researcher → Model evaluation: tests model
README.md - Researcher → Spectral extraction: extracts spectra
README.md - Spectral extraction → Cube utilities: loads cube utilities
h5_to_spectral_mean.py - Researcher → Spectral tools: inspects spectra
README.md