zym1119 / deeplabv3_mobilenetv2_pytorch Goto Github PK

可以给个cityscapes数据集下载的百度云链接吗？官网的我下载不下来

Hi，请问有已经训好的模型可以直接拿来用吗？

我用你的一模一样的网络和数据训练最后loss收敛到0.17

用提供的ImageNet_pretrain.pth预训练模型出现WARNING: Cannot load pre-trained model. Start training......
捕获error如下：
Error: Error(s) in loading state_dict for Sequential:
Missing key(s) in state_dict: "0.0.weight", "0.1.weight", "0.1.running_var", "0.1.running_mean", "0.1.bias", "1.block.0.weight", "1.block.1.weight", "1.block.1.running_var", "1.block.1.running_mean", "1.block.1.bias", "1.block.3.weight", "1.block.4.weight", "1.block.4.running_var", "1.block.4.running_mean", "1.block.4.bias", "1.block.6.weight", "1.block.7.weight", "1.block.7.running_var", "1.block.7.running_mean", "1.block.7.bias", "1.res_conv.0.weight", "1.res_conv.1.weight", "1.res_conv.1.running_var", "1.res_conv.1.running_mean", "1.res_conv.1.bias", "2.block.0.weight", "2.block.1.weight", "2.block.1.running_var", "2.block.1.running_mean", "2.block.1.bias", "2.block.3.weight", "2.block.4.weight", "2.block.4.running_var", "2.block.4.running_mean", "2.block.4.bias", "2.block.6.weight", "2.block.7.weight", "2.block.7.running_var", "2.block.7.running_mean", "2.block.7.bias", "3.block.0.weight", "3.block.1.weight", "3.block.1.running_var", "3.block.1.running_mean", "3.block.1.bias", "3.block.3.weight", "3.block.4.weigh

when I run the main.py.I get the problem.

Hello~ I am working on training the mobileNetV2+aspp these days. I found out if I only train the model on the cityscapes, the mIoU is very low.
Do you use the mobile net which is pretrained on the ImageNet?
Can you share the weigh of the mobile net V2 in pytorch?

I run your code,meet the error as flows:
square_avg.mul_(alpha).addcmul_(1 - alpha, grad, grad)
RuntimeError: cuda runtime error (59) : device-side assert triggered at /pytorch/aten/src/THC/generic/THCTensorMathPairwise.cu:81
could you help me?thanks!

方便的话可以加个VX吗

I meet this problem, can you help me ?

训练出现AttributeError: 'NoneType' object has no attribute 'min' 的错误。我现在用的是python3.6.6，不会版本问题吧。

@zym1119 你好，按照你代码的默认设置进行训练，训练后测试val的miou，我只能得到52%左右，mobilenet v2里面有75%左右，这差距也太大了吧，不知道你是不是测出来miou也很低？