Directly visualize the CDD data label and find a lot of noise

likyoo / Siam-NestedUNet

The pytorch implementation for "SNUNet-CD: A Densely Connected Siamese Network for Change Detection of VHR Images"

MIT License

260 stars 60 forks source link

Closed lsh1994 closed 3 years ago

lsh1994 commented 3 years ago

Does the unprocessed code affect the evaluation result?

参考代码：

from collections import Counter
out = cv2.imread(p_label, cv2.IMREAD_GRAYSCALE)
print(p_a,Counter(out.reshape(-1)))

(hope not to close other people's issues easily)

likyoo commented 3 years ago

你好，我按照你的方式打开并没有发现噪点。

lsh1994 commented 3 years ago

谢谢回复！Counter(out.reshape(-1)) 返回的结果怎样啊？查看下标签中的词频还是有不一样的值？

likyoo commented 3 years ago

很多数据集都有这样的情况，处理一下就可以