speedinghzl
Some questions about the evaluation of ccnet
Hi
I would like to ask why the evaluation result of ccnet source code will be 19 arrays, how do these 19 arrays represent the results of 500 proof sets?Another problem is that the ratio and number of training sets, validation sets, and test sets I use are the same for you. Why is my IU 51% and you are 80%?I am an undergraduate student and I urgently need your help. Thank you very much.
The 19 elements of arrays represent 19 categories. You are excepted to comprehend mean of IOU. Is your result obtained by carrying out run_local.sh without any modification?
你好，我注意到为什么你的代码中train和evaluate是分开进行的啊，我们训练不是应该训练完一次后马上进行评估吗，因为我是在windows中python编译器中进行的，没用run_local,我应该怎样做才能让这两个代码同时运行啊，求指教啊，停止不前了，需要你的帮助一下
@swjtulinxi 1.你的意思是训练过程中每个epoch都测试吧。这种操作不是必需的，速度会很慢。 2.在shell脚本里没法实现这个功能，需要在train.py里添加eval的逻辑。
AttributeError: 'NoneType' object has no attribute 'shape'
我也遇到了其它人说的问题，检查了几遍还是有问题
没有做模型筛选，直接取网络最后的保存的pth。
NoneType一般是由于提供给opencv的路径不对导致的。 请确保dataloader是可以访问cityscapes的图像，你可以打印一下dataloader读取的路径是否正确，并敲代码试一下是否可以读取图像数据。