nikopueringer

CorridorKey

AI#Machine Learning #Computer Vision#VFX#PyTorch#Python

// summary

CorridorKey is a neural network-based tool designed to solve the complex problem of unmixing foreground subjects from green or blue screen backgrounds. It reconstructs the true straight color and linear alpha channel for every pixel, effectively preserving fine details like hair and motion blur. The project supports high-fidelity VFX workflows by outputting 16-bit and 32-bit Linear float EXR files compatible with industry-standard compositing software.

// technical analysis

CorridorKey is a neural network-based green screen keying solution designed to solve the 'unmixing' problem, where traditional tools fail to accurately separate foreground subjects from background spill. By predicting the true straight color of the foreground and a clean linear alpha channel for every pixel, it preserves delicate details like motion blur and transparency that binary masks typically destroy. The project prioritizes high-fidelity VFX workflows by supporting 16-bit and 32-bit linear EXR outputs, offering a sophisticated alternative to manual rotoscoping and complex edge matte creation.

// key highlights

Physically accurate unmixing preserves hair, motion blur, and translucency by reconstructing the foreground object's true color.

Resolution-independent engine dynamically scales inference to handle 4K plates while maintaining a 2048x2048 high-fidelity backbone.

Native support for 16-bit and 32-bit Linear float EXR files ensures seamless integration into professional VFX pipelines like Nuke, Fusion, and Resolve.

Automated morphological cleanup system intelligently removes tracking markers and unwanted background artifacts from the final output.

Dedicated checkpoints for both green and blue screens allow for precise spill removal tailored to the specific background color used during filming.

Flexible hardware support includes NVIDIA CUDA, Apple Silicon via MLX, and AMD GPUs via ROCm, with optional modules for automated Alpha Hint generation.

// use cases

Physically accurate unmixing of foreground subjects from green or blue screens

Automatic generation of high-fidelity alpha channels and straight color passes

VFX-standard output for seamless integration into Nuke, Fusion, and Resolve

// getting started

To begin, clone the repository and use the provided installation scripts for your operating system, which leverage 'uv' to automatically manage Python and dependencies. Once installed, use the command-line wizard by dragging your video clips onto the launcher script to organize your files and generate necessary Alpha Hints. Finally, configure your settings within the wizard to process your clips and export the resulting foreground and matte files.