Expand description
GPU-accelerated video denoising operations.
This module provides compute-shader-based denoise filters for video frames. Three algorithms are available:
- Gaussian – simple spatial blur (low latency, lower quality)
NonLocalMeans– patch-based NLM denoising (higher quality, more compute)BilateralFilter– edge-preserving spatial filter (good quality/speed trade-off)
All operations fall back to CPU SIMD code when no suitable GPU is available
(the GpuDevice::new failure path returns an error that the caller may handle
by switching to a software path).
§GPU path (bilateral filter)
When device.is_fallback == false the bilateral filter is executed on the GPU
via a WGSL compute shader (shaders/bilateral.wgsl). The NLM algorithm is
too expensive for a single-dispatch compute shader (O(search_area² × patch_area)
per pixel) and is therefore kept as a CPU fallback; GPU NLM is planned for a
future release using multi-pass tiled reduction.
Structs§
- Denoise
Kernel - Denoise kernel configuration for use with the
kernelsmodule. - Denoise
Operation - GPU-accelerated denoise operations.
Enums§
- Denoise
Algorithm - Denoise algorithm variants.