AliceH1226的博客

训练了两回因为类的init函数已经调用了train函数

keaking caffe2 thread-pool after fork

pytorch生成绿色背景图片

default porcess group has not been initialized, please make sure to call init_process_syncbatchnorm 只能多GPU训练

报错信息expected scalar type Long but found Int或者expected scalar type Long but found Float报错场景pytorch的分类，本例具体为torch.nn.crossentropyloss原因pytorch的分类的loss函数的label的类型只能是torch.LongTensor类型(也是torch.int64类型），其他类型比如torch.int32，torch.float32等等会报错。我自己的是从np.long

x.repeat(m,n,z…)在第0维度把x重复m次，在第1维度重复重复m次，以此类推。import torchx = torch.tensor([1, 2, 3])y = x.repeat(4, 2)print(y)print(y.size())y = x.repeat(4, 2, 1)print(y)print(y.size())y = x.repeat(4,)print(y)print(y.size())tensor([[1, 2, 3, 1, 2, 3],

torch.cat()import torchb, c, w, h = 2, 3, 4, 4x = torch.zeros(b, c, w, h)print(x.shape) # torch.Size([2, 3, 4, 4])mean=[0.485, 0.456, 0.406] #rgbstd =[0.229, 0.224, 0.225]get_channel = x[:,[0]]print(get_channel.shape) # torch.Size([2, 1, 4, 4])