#158·Dolphin

code-element transcription produces hallucinated code in some cases

Author: Halo5082Created Nov 28, 2025Updated Nov 28, 2025

Hi Dolphin team — first, thank you for your hard work on this project. I found a case where the two-step process (layout parsing → per-element reading/transcription) yields incorrect output: the “code” element is sometimes full of hallucinated content that does not match the original image.

ISO IEC 14882 2017 bad_case.pdf

The error part in the output json: { "label": "code", "bbox": [ 87, 115, 531, 739 ], "text": "#!/usr/bin/env python\n#PLICIT constexpr tuple(tuple<UTypes...>&k);\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env python\n#!/usr/bin/env pytho ...... /env\nbin", "reading_order": 1 },