Crate oxigdal_ml_foundation 

OxiGDAL ML Foundation

Deep learning training infrastructure and model architectures for geospatial machine learning.

This crate provides:

• Training infrastructure (loops, optimizers, schedulers, losses)
• Transfer learning and fine-tuning capabilities
• Model architectures (UNet, ResNet, custom layers)
• Data augmentation pipelines for geospatial imagery
• Evaluation metrics and monitoring

Features

• std (default): Standard library support
• pytorch: PyTorch backend for training (requires libtorch)
• onnx: ONNX export support for trained models
• cuda: GPU acceleration (requires CUDA)

Architecture

The crate is organized into several modules:

• training: Training loops, optimizers, schedulers, and losses
• transfer: Transfer learning and fine-tuning
• models: Neural network architectures
• augmentation: Data augmentation pipelines
• data: Data pipeline with dataset loaders and batching
• metrics: Evaluation metrics

COOLJAPAN Compliance

• Pure Rust implementation (PyTorch bindings are feature-gated)
• No unwrap() calls in production code
• All files under 2000 lines
• Uses workspace dependencies
• Uses SciRS2-Core for numerical operations (Pure Rust Policy)

Example

use oxigdal_ml_foundation::{
    training::TrainingConfig,
    training::training_loop::Trainer,
    models::unet::UNet,
    metrics::Metrics,
};

// Create a UNet model for segmentation (in_channels=3, num_classes=10, depth=4)
let model = UNet::new(3, 10, 4)?;

// Configure training
let config = TrainingConfig {
    learning_rate: 0.001,
    batch_size: 16,
    num_epochs: 100,
    ..Default::default()
};

// Train the model (requires PyTorch feature)
// let trainer = Trainer::new(model, config)?;
// let trained_model = trainer.train(train_data, val_data)?;

Re-exports

pub use error::Error;

pub use error::Result;

Modules

augmentation

Data augmentation pipelines for geospatial imagery.

data

Data pipeline for geospatial machine learning.

error

Error types for ML foundation operations.

metrics

Evaluation metrics for machine learning models.

models

Neural network model architectures.

training

Training infrastructure for deep learning models.

transfer

Transfer learning and fine-tuning capabilities.

Constants

NAME

Library name

VERSION

Library version

Functions

has_gpu_support

Checks if GPU support is available

has_onnx_export

Checks if ONNX export is available

has_pytorch_backend

Checks if PyTorch backend is available