Deblur Image (Restormer)

@swz301 VersionEnhance

Transformer restorer for an Image that runs a per-task checkpoint — motion or defocus deblur, real / Gaussian denoise, or deraining — and emits a restored RGB Image at the original resolution. Use when the degradation matches one of the task variants.

How it fits

Typical backends

An HTTP restoration service: degraded frames are run through the task checkpoint and returned over HTTP.

Pre-OCR cleanup: a deraining or denoise pass removes weather / sensor artefacts from a captured frame before text recognition.

Caveats

Hard constrainttask selects the network's internal channel count and normalisation and must match the bound model checkpoint; any value outside the supported task set aborts startup.
I/O contractThe single-channel Gaussian-grayscale task converts the input to luma before inference and broadcasts the result back across three channels, so its output is monochrome even though it is tagged RGB.
CostWhole-image inference is a single forward pass that can exhaust GPU memory on large frames; enabling patch-wise tiling switches to overlapped patch inference that bounds VRAM, with adjacent patches blended to avoid seams.
Hard constraintThe `.pth` checkpoint is read from the bound model weights slot and must match task; switching task or the bound weights requires a restart.
FallbackA `cuda` device falls back to CPU when no GPU is present, which is impractical for live video; an empty input Image is returned unchanged without invoking the network.
LimitationInput is converted to RGB and output is always RGB at the input height and width — the component restores but does not super-resolve.

Versions

33efd39ddefaultlatest
Automated release
6/27/2026