ContextualConv

Flexible, global-context conditioned convolutional layers for PyTorch. Supports:

• Per-channel bias (original behavior)

• Per-channel scale

• Full FiLM-style modulation (scale + bias)

Contents:

• Usage Guide
  • 🔧 1D Example (no context)
  • 🧠 1D Example with context (bias only)
  • 🧠 1D Example with context (scale only)
  • 🔁 1D Example with FiLM (scale + bias)
  • 🧠 1D with MLP for context
  • 🧠 Multi-layer MLP for context
  • 🖼️ 2D Example with FiLM-style context
  • ✅ Notes
• How It Works
  • 🧠 ContextualConv Layers
  • 🔄 Forward Pass
  • 🧩 Architecture (1D/2D)
  • ✅ Notes
• FAQ
  • ❓ What is the context vector used for?
  • ❓ How is the context processed?
  • ❓ Does the context affect each spatial location differently?
  • ❓ What happens if I don’t provide context_dim?
  • ❓ Can this replace SE blocks or FiLM?
  • ❓ Is it compatible with grouped/depthwise convolutions?
• Architecture Overview
• ContextualConv
  • Usage Guide
  • How It Works
  • FAQ
  • Architecture Overview
• API Reference