“I think the actual paper (linked at the end) is much more informative and interesting than this coverage: Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks”
Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks
单个深度卷积网络直接读街景门牌号,reCAPTCHA 达 99.8%
提出一个统一的方法,用一个深度卷积网络直接在像素上同时完成定位、分割和识别,而不是分步处理,并用 DistBelief 训练大规模分布式网络;性能随网络加深而提升,最佳为 11 个隐藏层。在 SVHN 上整串门牌号识别准确率超过 96%,单数字准确率 97.84%;在更难的街景数据上超过 90%;在合成的最难 reCAPTCHA 上达到 99.8%。
评论者主要引用它的 99.8% 这一数字,说明神经网络解 reCAPTCHA 比人更强,也把它当作 Google 街景门牌号识别系统的细节来源。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“and it links to the exact same article on arxiv, titled:”
“It seems to be actually 99.8% accuracy according to their paper: Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks”
“We found that neural networks can solve CAPTCHAS much better than humans, 99.8% on the "hard" ReCAPTCHA instances: Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks”
“99.8% accuracy in solving distorted text reCAPTCHA”
“Sure, there are a lot of potential things to do - either by enumerating towards the full "brute force" loss partway as I think you are hinting toward, or treating as a subsequences, or something else [0, 1, 2].”
看全部 7 条 →
“The state of the art has improved so much that for the older tasks, computers are arguably better than humans at them:”