Some neural networks will be mentioned in ImageNet Top-5 or Top-1, which is a standard picture detection accuracy rate, before the introduction of the first to introduce ImageNet.
【】 ImageNet
ImageNet program is used to retrieve a large object visual object recognition database. As of 2016, ImageNet already more than ten million manually annotated images, category labeled images. At least one million images also provide a bounding box.
Since 2010, ImageNet software organizes the annual contest, called (ImageNet Large Scale Visual Recognition Challenge, ILSVRC). The main content is to achieve correct classification and detection algorithms to identify objects and scenes through the program, the evaluation criteria is the Top-5 error rate.
-
Top-5 error rate
That is a picture, if the probability of five contains the correct answer, which is considered correct.
-
Top-1 error rate
That is a picture, if the most probable answer is correct, it thinks is right.
Original: https://blog.csdn.net/v1_vivian/article/details/73251187