zhazhaming

恩，cityscapes有点大，我只用了gtFine的都有60G，我试试把他传到网盘吧

linux没有网盘客户端，最多一次300个，我正在一个城市一个城市传图片，可能要等几天才好

https://blog.csdn.net/zym19941119/article/details/81198315 我传好了cityscapes的数据集到百度云，连接在我的博客里，还有kitti的也传了

最近在实习，没时间改，本地的代码在60多，具体多少忘了，很久没看了，准备等秋招结束再整理更新

this is because the image is not read by cv2,imread, have you set the right path to dataset_root?

are you load the epoch_100 checkpoint? this was a early version and some changes have been made so if you load this checkpoint you may fail. if you mean epoch_150...

sorry to reply you so late, can you show me more error information plz?

i thought this may be caused by wrong in label tensor, maybe your label have negative value like -1, in cityscapes, the trainid of 'license plate' is -1, this may...

Sorry I dont have the imagenet pre-trained weight, cause I dont have ImageNet in my computer. I got a little busy recently :-) , I will try to get others...

i upload a ImageNet pre-trained weight, you can try it :)