Unwarp Document (docTR)

@haofeng1 VersionEnhance

Rectifies a heavily curled or perspective-distorted document Image with a segmentation mask plus the GeoTr transformer, which predicts a backward map that is upsampled and applied to emit a flattened Image for OCR.

How it fits

Typical backends

Flatten a curled page before OCR: a file source feeds the image, this component rectifies it with the transformer unwarper, and a recognizer reads the cleaned text out as JSON.

Rectify documents from a live HTTP feed and publish the flattened image for visual inspection instead of OCR.

Caveats

AccuracyThe transformer predicts a small backward map at a fixed working resolution and upsamples it, so very high input resolutions do not recover finer text detail; hands, torn edges, multiple folded panels, and non-document inputs remain common failure cases.
I/O contractAn image smaller than two pixels on a side is returned unchanged; otherwise the output is a same-color-space rectified Image.
LatencyEach frame runs a U2Net segmentation pass and the GeoTr transformer, so a GPU is strongly recommended; CPU execution works but is markedly slower and impractical for live throughput.
LimitationThis is a dewarping-only restorer with no deshadow, deblur, or binarize modes; pair it with a dedicated cleanup stage if illumination correction is also required.

Versions

895b771blatestdefault
Automated release
6/22/2026