ununennium 1.0.5

Production-grade Python library for satellite and geospatial imagery machine learning

Ununennium

Production-grade Python library for satellite and geospatial imagery machine learning.

Documentation • PyPI • GitHub • Examples

---

Overview

Ununennium (Element 119, the next alkali metal) represents the cutting edge of satellite imagery machine learning. This library provides a unified, GPU-first framework for end-to-end Earth observation workflows, from cloud-native data access through model training to deployment.

Why Ununennium?

Challenge	Traditional Approach	Ununennium Solution
CRS Handling	Manual, error-prone	Automatic CRS tracking with GeoTensor
Large Rasters	Memory overflow	Streaming I/O with COG/Zarr
Multi-spectral	Custom band handling	First-class n-band support
Reproducibility	Ad-hoc seeds	Deterministic training pipeline
Physics	Purely data-driven	Physics-informed constraints (PINN)

---

Key Features

Module	Capability	Highlight
core	GeoTensor, GeoBatch	CRS-aware tensors with automatic coordinate tracking
io	COG, STAC, Zarr	Cloud-native streaming with lazy loading
models	CNN, ViT, GAN, PINN	15+ architectures with registry pattern
training	Trainer, Callbacks	Mixed precision, gradient accumulation, DDP
preprocessing	Indices, Normalization	NDVI, EVI, SAVI with sensor-aware math
augmentation	Geometric, Radiometric	CRS-preserving transforms
tiling	Sampler, Tiler	Overlap-aware patch extraction
metrics	IoU, Dice, ECE	Calibrated uncertainty quantification
export	ONNX, TorchScript	Production deployment ready

---

Performance Benchmarks

Benchmarks on NVIDIA A100 80GB, PyTorch 2.1, CUDA 12.1:

Model	Input Size	Batch	Throughput	Memory	mIoU
U-Net ResNet-50	512×512×12	16	142 img/s	12.4 GB	0.78
U-Net EfficientNet-B4	512×512×12	16	98 img/s	14.2 GB	0.81
ViT-L/16	224×224×12	32	256 img/s	18.1 GB	0.83
Pix2Pix	256×256×12	8	67 img/s	8.6 GB	N/A

---

Installation

# Core installation
pip install ununennium

# With geospatial dependencies (rasterio, pyproj, shapely)
pip install "ununennium[geo]"

# Full installation with all features
pip install "ununennium[all]"

Requirements

• Python 3.10+
• PyTorch 2.0+
• NumPy 1.24+
• GDAL 3.4+ (optional, for geospatial I/O)

---

Quick Start

Load Satellite Imagery

import ununennium as uu

# Read with automatic CRS detection
tensor = uu.io.read_geotiff("sentinel2_l2a.tif")
print(f"Shape: {tensor.shape}")     # (12, 10980, 10980)
print(f"CRS: {tensor.crs}")         # EPSG:32632
print(f"Resolution: {tensor.resolution}")  # (10.0, 10.0)

Train a Segmentation Model

from ununennium.models import create_model
from ununennium.training import Trainer, CheckpointCallback
from ununennium.losses import DiceLoss
import torch

# Create U-Net with ResNet-50 backbone
model = create_model(
    "unet_resnet50",
    in_channels=12,      # Sentinel-2 bands
    num_classes=10,      # Land cover classes
)

# Configure training
trainer = Trainer(
    model=model,
    optimizer=torch.optim.AdamW(model.parameters(), lr=1e-4),
    loss_fn=DiceLoss(),
    train_loader=train_loader,
    val_loader=val_loader,
    callbacks=[CheckpointCallback("checkpoints/")],
    mixed_precision=True,
)

# Train with progress tracking
history = trainer.fit(epochs=100)

Physics-Informed Learning

from ununennium.models.pinn import PINN, DiffusionEquation, MLP

# Define PDE constraint
equation = DiffusionEquation(diffusivity=0.1)

# Create PINN
pinn = PINN(
    network=MLP([2, 128, 128, 1]),
    equation=equation,
    lambda_pde=10.0,  # Weight for physics loss
)

# Train with collocation points
losses = pinn.compute_loss(x_data, u_data, x_collocation)

---

Architecture

ununennium/
├── core/           # GeoTensor, GeoBatch, types, CRS handling
├── io/             # COG, STAC, Zarr readers/writers
├── preprocessing/  # Normalization, spectral indices
├── augmentation/   # Geometric and radiometric transforms
├── tiling/         # Spatial sampling and tiling
├── datasets/       # Dataset abstractions
├── models/         # Backbones, heads, GAN, PINN
│   ├── backbones/  # ResNet, EfficientNet, ViT
│   ├── heads/      # Classification, Segmentation, Detection
│   ├── gan/        # Pix2Pix, CycleGAN, ESRGAN
│   └── pinn/       # Physics-informed networks
├── losses/         # Dice, Focal, Perceptual, Physics
├── metrics/        # IoU, Dice, Calibration
├── training/       # Trainer, Callbacks, Distributed
├── export/         # ONNX, TorchScript
└── sensors/        # Sentinel-2, Landsat, MODIS specs

---

Supported Tasks

Task	Models	Metrics
Scene Classification	ResNet, EfficientNet, ViT	Accuracy, F1, AUC
Semantic Segmentation	U-Net, DeepLabV3, FPN	mIoU, Dice, PA
Object Detection	Coming Soon	mAP, AP50
Change Detection	Siamese + Diff	F1, κ
Super-Resolution	ESRGAN, Real-ESRGAN	PSNR, SSIM, LPIPS
Image Translation	Pix2Pix, CycleGAN	FID, SAM
Physics-Informed	PINN	L2 Error, PDE Residual

---

Documentation

• Getting Started
• API Reference
• Tutorials
• Model Zoo
• Theory

---

Authors

• Olaf Yunus Laitinen Imanov - Lead Developer & Architect
• Hafiz Rzazade - Core Contributor
• Laman Mamedova - Documentation & Testing
• Farid Mirzaliyev - Model Development
• Aian Ajili - Infrastructure & CI/CD

---

Citation

If you use Ununennium in your research, please cite:

@software{ununennium2024,
  title = {Ununennium: Production-grade Satellite Imagery Machine Learning},
  author = {Laitinen Imanov, Olaf Yunus and Rzazade, Hafiz and Mamedova, Laman and Mirzaliyev, Farid and Ajili, Ayan},
  year = {2024},
  version = {1.0.0},
  url = {https://github.com/olaflaitinen/ununennium}
}

---

License

Apache License 2.0. See LICENSE for details.

---

Contributing

We welcome contributions! Please read CONTRIBUTING.md for guidelines.

Development Setup

git clone https://github.com/olaflaitinen/ununennium.git
cd ununennium
pip install -e ".[dev]"
pre-commit install
pytest tests/

---

Built with passion for Earth observation and machine learning.

Ununennium: Where geospatial meets deep learning.