Unverified50% confidenceFactExact time
By around 2013, deep learning-based OCR technology had already achieved a recognition rate of over 90% on text-distortion-type CAPTCHAs.
1
Sources
50%
Confidence
Long-term
Relevance
9/10/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2012年前后基于深度学习的OCR已能以超过99%的准确率破解reCAPTCHA v178% similarUnverified深度学习时代的LSTM和CNN架构大幅提升了OCR识别准确率，但真实业务场景中端到端准确率仍难以达到100%70% similarUnverifiedIn the deep learning era, end-to-end models based on CNN and RNN (such as the CRNN architecture) significantly improved OCR recognition capabilities, and in recent years the Transformer architecture has also been introduced into the OCR field.66% similarUnverified内容感知裁剪结合了计算机视觉中的显著性检测和目标识别技术，现代深度学习方案如 U²-Net、DeepGaze 使用卷积神经网络学习视觉焦点64% similarUnverifiedDeepSeek has boosted the inference speed of its frontier AI by up to 85% through speculative decoding technology63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/885908API
curl https://kongchang.com/api/v1/knowledge/claims/885908MCP
get_claim(id=885908)