SAGA Documentation

Welcome to the documentation for SAGA (Spatially-Adaptive Gated Activation).

SAGA is an optimized, plug-and-play activation function featuring fused Triton kernels and dynamic gate map extraction capabilities for general deep vision models. Originally developed for medical image restoration, SAGA v0.2.0 is designed for high-speed, interpretable deep learning across all image processing tasks.

Contents:

• Installation Guide
  • Prerequisites
  • Installing via PyPI
  • Installing from Source
  • Verifying the Installation
• Mathematical Background
  • SAGA formulation
  • Parameter count overhead
• API Reference
  • saga.activation
  • saga.blocks
  • saga.utils
• Performance and Visual Gallery
  • Visual Comparison: Best-in-Class Models
  • Natural Image Dehazing
  • Quantitative Benchmarks
  • Computational Efficiency & Hardware Acceleration
  • Citation

Indices and tables

• Index

• Module Index

• Search Page