如何应用在YOLOv3上？

[qhykwsw]

你好，我在使用你的代码的时候，有点困惑。得到的结果如下： Accuracy: 87.11% Boxes: [[0.0234375 0.06041667] [0.02109375 0.046875 ] [0.01875 0.04166667] [0.028125 0.05104167] [0.0140625 0.028125 ] [0.0421875 0.08854167] [0.009375 0.02083333] [0.02109375 0.04166667] [0.0328125 0.06979167]] Ratios: [0.39, 0.45, 0.45, 0.45, 0.47, 0.48, 0.5, 0.51, 0.55]

结果是相对比例，但是YOLOv3是使用的绝对值。我的想法是应该对每个值乘上输入进网络的尺寸，比如输入网络尺寸是608x608。那anchor大小尺寸就是上面的每一个值乘以608。请问我的理解没错吧？

[lars76]

Hi, yeah this is how I used the boxes too. However, in the original paper the authors apparently not only multiplied the boxes by 608 but also used different scaling factors. But I don't think that this scaling step is really that necessary. Especially because the authors didn't write what scaling factors they used exactly.

If the objects can't vary a lot between different datasets, I would just use the dimension of the image (e.g 608x608).

[AnkyZhang]

你好，请问如何确定哪三个是scale1的anchor boxes，哪三个是scale2的，哪三个是scale3的呢？

[sisrfeng]

你好，请问如何确定哪三个是scale1的anchor boxes，哪三个是scale2的，哪三个是scale3的呢？

请问你解决这个问题了吗？

[Andy-zhujunwen]

你好，我在使用你的代码的时候，有点困惑。得到的结果如下： Accuracy: 87.11% Boxes: [[0.0234375 0.06041667] [0.02109375 0.046875 ] [0.01875 0.04166667] [0.028125 0.05104167] [0.0140625 0.028125 ] [0.0421875 0.08854167] [0.009375 0.02083333] [0.02109375 0.04166667] [0.0328125 0.06979167]] Ratios: [0.39, 0.45, 0.45, 0.45, 0.47, 0.48, 0.5, 0.51, 0.55]

结果是相对比例，但是YOLOv3是使用的绝对值。我的想法是应该对每个值乘上输入进网络的尺寸，比如输入网络尺寸是608x608。那anchor大小尺寸就是上面的每一个值乘以608。请问我的理解没错吧？

你好，为什么我运行代码后都没输出啊？输出如下： (base) E:\datasets\kmeans-anchor-boxes-master\kmeans-anchor-boxes-master>python -m unittest discover -s tests/ ..............

Ran 14 tests in 18.978s

OK

没有box什么的。。为什么啊。。。

[sisrfeng]

这个好像可以参考https://colab.research.google.com/drive/1sQv8QguVMnhkaaEvlOZRdwVJk5C1jkKo#scrollTo=_URrqg10_iMW