CHAOS-Bench
A benchmark specifically designed to evaluate OCR models' ability to handle anomalous text, using Micro Recall as the primary evaluation metric.
Timeline (last 90 days)
On the Qwen3.5-2B model, GAD-RL achieved a Micro Recall of 59.92% on CHAOS-Bench, surpassing GRPO and fixed-weight GRPO+OPD by 8.45 and 4.43 percentage points respectively
CHAOS-Bench is a benchmark specifically designed for OCR faithfulness evaluation, containing counter-intuitive text such as spelling errors, abbreviations, numeric strings, and mixed symbols
All Facts (2)
On the Qwen3.5-2B model, GAD-RL achieved a Micro Recall of 59.92% on CHAOS-Bench, surpassing GRPO and fixed-weight GRPO+OPD by 8.45 and 4.43 percentage points respectively
50%UnverifiedCHAOS-Bench is a benchmark specifically designed for OCR faithfulness evaluation, containing counter-intuitive text such as spelling errors, abbreviations, numeric strings, and mixed symbols
50%