Current Status
AUCMEDI is still under active development, which is why the official status is:
Work in Progress
Roadmapยค
Current Progress:
Features that are already supported by AUCMEDI:
- Binary, multi-class and multi-label image classification
- Support for 2D as well as 3D data
- Handling class imbalance through class weights & loss weighting like Focal loss
- Stratified iterative sampling like percentage split and k-fold cross-validation
- Standard preprocessing functions like Padding, Resizing, Cropping, Normalization
- Extensive online image augmentation
- Automated data loading and batch generation
- Data IO interfaces for csv and subdirectory encoded datasets
- Transfer Learning on ImageNet weights
- Large library of popular modern deep convolutional neural network architectures
- Ensemble Learning techniques like Inference Augmenting
- Explainable AI (XAI) via Grad-Cam, Backpropagation, ...
- Clean implementation of the state-of-the-art for competitive application like challenges
- Full (and automatic) documentation of the complete API reference
- Started creating examples & applications for the community
- Available from PyPI for simple installation in various environments
- Interface for metadata / pandas or NumPy table inclusion in model architectures
- Unittesting -> CI/CD
- Clean up Website
- Integration of bagging and stacking pipelines for utilizing ensemble learning techniques
- Integrate evaluation functions
- Support for AutoML via CLI and Docker
- Documentation for AutoML
Planed milestones and features are:
- Examples
- Tutorials
- Publication