TextVerifier: Seeing Beyond Recognition to Verify Chinese Text Rendering

Key points

• Such errors are invisible to OCR and vision-language models, which are trained for recognition and are therefore invariant to the very structural deviations that define a rendering error.1
• We formalize text rendering verification as a target-conditioned problem: given a target character and an image crop, decide whether the crop is a correct rendering of that character.1
• Our verifier TEXTVERIFIER feeds the text tower a character's Ideographic Description Sequence (IDS) rather than a single-character symbol, and is trained on 12.7M crop-IDS pairs plus label-free "same-character, different-quality" hard pairs sampled from the training trajectory of a text-region inpainting model.1
• To evaluate this ability we build TEXTRENDERBENCH-ZH, the first per-character benchmark for generated Chinese text, annotating position, identity, and correctness for 7,358 characters.1

Sources (2)

• [1]TextVerifier: Seeing Beyond Recognition to Verify Chinese Text Rendering
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 04:07 AM

  Text-to-image models increasingly generate readable Chinese text that still contains local rendering defects, such as a character with an extra or missing stroke or a malformed component.

  Such errors are invisible to OCR and vision-language models, which are trained for recognition and are therefore invariant to the very structural deviations that define a rendering error.

• [2]TextVerifier: Seeing Beyond Recognition to Verify Chinese Text Rendering
  Hugging Face Daily Papers · Oct 14, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.