FlyOCR
FlyOCR
Choose a recorded PDF example, play or scrub the inference trace, and inspect the raw recognition results.
Runs on your device. A desktop browser is recommended for the full controls.
Built from the credited original source. Runs on your device in a separate Flymaxxing runtime.Loading the project… Large models may need a few minutes on their first launch.
About this version
Ten original recorded experiments. This viewer does not perform fresh OCR on an uploaded PDF. Source + adaptation
About FlyOCR
Give the fly a PDF. This experiment turns individual printed characters into neural activity and reads them back out as text.
Model & implementation
Retains the MaleCNS graph with frozen internal weights and a trained decoder. The README reports 87.6% on a 1,632-glyph benchmark and 5.7% character error on 176 reference characters. These are limited-font, selected-crop experiments. The 3D fly is illustrative animation beside recorded OCR.
Inspect the source without leaving Flymaxxing
Pinned source revision · loaded from the original public repository
Choose a file to read its source.