MIT的《深度学习》精读（14）

最新推荐文章于 2024-09-18 20:39:31 发布

caimouse

最新推荐文章于 2024-09-18 20:39:31 发布

阅读量1.1k

点赞数

分类专栏：深度学习文章标签： ai python tensorflow 人工智能深度学习

本文链接：https://blog.csdn.net/caimouse/article/details/68067576

版权

深度学习专栏收录该内容

127 篇文章 15 订阅

订阅专栏

However, for many tasks, it is difficult to know what features should be extracted. For example, suppose that we would like to write a program to detect cars in photographs. We know that cars have wheels, so we might like to use the presence of a wheel as a feature. Unfortunately, it is difficult to describe exactly what a wheel looks like in terms of pixel values. A wheel has a simple geometric shape but its image may be complicated by shadows falling on the wheel, the sun glaring off the metal parts of the wheel, the fender of the car or an object in the foreground obscuring part of the wheel, and so on.

但是，在许多人工智能的任务中根本无法识别那些特征需要提取出来。比如，我们需要开发一个软件从图片里识别汽车。按照常识来说，我们知道汽车有轮子，所以会拿是否有轮子来当作特征值。不幸的是，我们很困难地描述轮子是什么样子。虽然轮子在几何上看来是有一定的形状，但是这些轮子可能被一些阴影覆盖着，或者轮毂被太阳光照耀着，或者汽车的档泥板阻挡住一部分，又或者车前头部分掩盖着一部分，以上种种原因很难描述出图片里的轮子是什么样子。