关于config文件里的n_cats设置

[yangaiping]

我通过check_dataset_info.py脚本查到我的所有数据集标签图像像素值区间是[0，132]，但是我的数据集标签类别一共才13类，为什么n_cats设成13我的训练就会报错呢，但是设成133又能成功运行？

[CoinCheung]

Your category does not in range of [0, 13], but in [0, 131] instead. You should convert the label values into [0, 13].

[yangaiping]

您是什么意思呢，我的图像像素值区间是[0，132]，但是我的像素颜色种类是13，那我的n_cats应该设为多少呢

[CoinCheung]

You should map the pixel value of your label file into range of [0, 12].

[yangaiping]

代码里的这个输出[0，132]难道不是像素的最小值与最大值形成的区间吗

[CoinCheung]

You have 13 categories, but your category index are not continously ranging from 0 to 12, they are sparsely ranging from 0 to 132.

You should map your raw label values continuously into range [0, 12].

For example, in the picture above, we can see the mapping rule as:

0    -> 0
14  -> 1
19  -> 2
....

You should process your label files to by-pass those empty categories, or you have to set the number of categories into 132.

[CoinCheung]

They are expected to be singe channel image.

[yangaiping]

我明白您的意思了，我的png格式的标签文件有问题，我应该下去修改，是需要把彩色标签图像改成单通道8位标签图像吗

[CoinCheung]

Yes, they are expected to be single channel with label values rather than pixel values.

[yangaiping]

非常感谢您，我再下去试试。再次感谢