pytorch代码调试

1. 是否全部用的pytorch内置函数

2.高斯核参数均值μ和sigma是否设置合理

3.内存报错：batch size, pin_memory=False

4. pip install torchsnooper  在函数前加装饰器

@torchsnooper.snoop()

可以打印该函数的每行代码输出内容。

5.TypeError: add received an invalid combination of arguments - got (torch.cuda.FloatTensor), but expected one of: * (float value)

可能是有些tensor在GPU，有些在CPU

6.AttributeError: 'torch.Size' object has no attribute 'size'              torch0.1.12版本

torch.expand_as(x.size) →torch.expand(x.size)

7.torch.load    'utf-8' codec can't decode byte 0xae in position 3:   

Solutions: 

loaded_state_dict = torch.load('./runs1/Resnet50/model_best.pth.tar')

→   loaded_state_dict = torch.load('./runs1/Resnet50/model_best.pth.tar', encoding='latin1')

....

持续更新