hyperbench 0.1.0

Synthetic benchmarking for hyperspectral super-resolution methods

HyperBench

HyperBench is a synthetic benchmarking framework for fusion based hyperspectral super-resolution (HSR) methods. It provides a standardized, reproducible pipeline for evaluating models using controlled degradations, consistent metrics, and structured experiment outputs.

HyperBench is designed as an inference-time evaluation tool. It does not perform model training. Instead, it enables users to test already-trained models under a wide range of synthetic scenarios.

Project Home: https://github.com/ritikgshah/HyperBench. Please visit the github repository for exact implementation details and documentation. This PyPi project is solely meant to serve as the distribution method.

---

Key Features

Synthetic Degradation Pipeline

HyperBench generates realistic low-resolution inputs from high-resolution hyperspectral scenes:

• Spatial degradation via configurable Point Spread Functions (PSFs)
• Spectral degradation via configurable Spectral Response Functions (SRFs)
• Controlled noise injection using SNR (dB)

---

Flexible Experiment Design

Users can define experiments using:

• Explicit degradation specifications
• PSF configurations (type, sigma, kernel size)
• Noise levels (spatial and spectral)
• MSI band counts

---

Model-Agnostic Adapter Interface

HyperBench works with any model that follows a simple contract:

def run_pipeline(HR_MSI, LR_HSI, srf, psf=None, metadata=None): return prediction

or

return prediction, stats

Supported formats: - NumPy - TensorFlow - PyTorch

---

Built-in Metrics

• RMSE
• PSNR
• SSIM
• UIQI
• ERGAS
• SAM

---

Output Clipping Policy

Optional clipping to [0,1] before metric evaluation.

---

Structured CSV Logging

Each experiment produces structured CSV output with: - parameters - metrics - optional model stats

---

Visualization Utilities

• RGB visualization
• band inspection
• PSF and SRF visualization
• spectral plots

---

Framework Support

• NumPy
• TensorFlow
• PyTorch

Automatic device handling included.

---

Installation

pip install hyperbench

---

Command Line Interface

hyperbench run
--config config.yaml
--pipeline-module path/to/model.py
--method-name MyModel
--input-backend tensorflow

---

Configuration

Supports JSON and YAML configuration files.

---

Design Philosophy

• Reproducibility
• Standardization
• Flexibility

---

Scope

HyperBench is strictly an evaluation framework.

---

License

MIT License