评论
打印
If you've ever wondered what makes the perfect selfie, some of the world's most powerful computers have been pondering the same question.
如果你在琢磨怎样自拍更好看，世界上最强大的电脑也在思考着同样的问题。
Researchers have trained a neural network designed to think like the human brain to analyse millions of selfies to find what works.
研究者训练设计的神经网络像人类那样思考，以分析数百万张自拍照的奥秘。
They says the main key is to be female - and to show your hair, but cut off your forehead in your snaps.
他们得出的结果是，照片中的你首先要是个女性，——你要秀出你的头发，把额头部分裁掉。
Andrej Karpathy, a PhD student at Stanford working on Deep Learning, decided to use a powerful, 140-million-parameter state-of-the-art Convolutional Neural Network to help work out why some selfies are more popular than other.
安德雷·卡帕西，斯坦福一个研究计算机深度学习的博士生，决定使用包含1.4亿个参数的强大卷积神经网络来研究为何有的自拍更受欢迎。
'We decided to feed it 2 million selfies from the internet, and train it to classify good selfies from bad ones,' he explained in a blog.
他在博客中这样解释：“我们决定使用互联网上的两百万张照片，教会系统区分出好照片。”
The neural network, known as ConvNet 'looked' at every one of the 2 million selfies several tens of times.
这个叫做“ConvNet”的神经网络系统，要“看”两百万照片中的每一张数十次。
Each time, it tuned its filters in a way that best allows it to separate good selfies from bad ones.
每一次，它都要调整过滤器，以最好地实现好坏照片的区分。
'We can't very easily inspect exactly what it found (it's all jumbled up in 140 million numbers that together define the filters),' said Karpathy.
卡帕西说，“要检查系统到底发现了什么并不容易，（1.4亿个参数混杂在一起定义了过滤器）。”
'However, we can set it loose on selfies that it has never seen before and try to understand what it's doing by looking at which images it likes and which ones it does not. '
“但是我们可以在系统没审阅过的自拍图片上放松系统，试图通过看系统对图片的好恶看出些端倪。”
'A few patterns stand out for me,' he said.
“慢慢的就发现了一些规律。”他说。
Women were consistently ranked higher than men - to the extent there was not a single male in the top 100.
女性的自拍总是能评上高分——以至于前100名连一个男性也没有。
The position and pose of the face is quite consistent among the top images.
排名靠前的图片中，脸部的拍摄位置和姿态都很一致。
The face always occupies about 1/3 of the image, is slightly tilted, and is positioned in the center and at the top.
图片的1/3是脸部，略倾斜，位于中央或图片上部。
