OSV_q AttributeError: ‘numpy.ndarray‘ object has no attribute ‘clone‘

终于进入训练过程了

Load Data: 11
====>>   Build Net

====>>   Load I model

====>>   Train
D:\ProgramData\Anaconda3\envs\python36\lib\site-packages\torch\nn\modules\loss.py:446: UserWarning: Using a target size (torch.Size([1, 1, 3, 320, 320])) that is different to the input size (torch.Size([1, 3, 320, 320])). This will likely lead to incorrect results due to broadcasting. Please ensure they have the same size.
  return F.mse_loss(input, target, reduction=self.reduction)

先解决这个

之前为了让几个数值的维度统一，给b2参数也加了维度，采用debug测试发现，维度就是那个和别人都不一样的维度，所以把这行代码注释了就好了

b2 = b2.transpose(2, 0, 1)
b2 = b2[np.newaxis,:,:,:]   # 给数组增加维度
b2 = np_to_torch(b2).to(device)

==>> Thu Mar  3 16:29:20 2022 [Iteration 1 Epoch 50]    Loss = 19851.923828
==>> Thu Mar  3 16:29:21 2022 [Iteration 1 Epoch 100]    Loss = 10989.599609
==>> Thu Mar  3 16:29:23 2022 [Iteration 1 Epoch 150]    Loss = 7369.253418
==>> Thu Mar  3 16:29:25 2022 [Iteration 1 Epoch 200]    Loss = 8838.803711
==>> Thu Mar  3 16:29:26 2022 [Iteration 1 Epoch 250]    Loss = 10214.323242
==>> Thu Mar  3 16:29:28 2022 [Iteration 1 Epoch 300]    Loss = 8534.051758
==>> Thu Mar  3 16:29:29 2022 [Iteration 1 Epoch 350]    Loss = 8817.697266
==>> Thu Mar  3 16:29:31 2022 [Iteration 1 Epoch 400]    Loss = 8884.304688
==>> Thu Mar  3 16:29:32 2022 [Iteration 1 Epoch 450]    Loss = 9845.090820
==>> Thu Mar  3 16:29:34 2022 [Iteration 1 Epoch 500]    Loss = 8570.036133
==>> Thu Mar  3 16:29:34 2022 [Iter 1/8]   loss = 8570.036133
====>> Thu Mar  3 16:29:34 2022   Pass time: 0:00:17.553348
Traceback (most recent call last):
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 358, in <module>
    train(opt)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 284, in train
    grd_x, grd_y = getGrd(img_in)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 37, in getGrd
    grd_x = data_in.clone()
AttributeError: 'numpy.ndarray' object has no attribute 'clone'

又报错

这个问题真的找不出来，用debug报错是这样的，明天继续改

==>> Sun Mar  6 18:24:10 2022 [Iter 1/8]   loss = 7767.667969
====>> Sun Mar  6 18:24:10 2022   Pass time: 0:00:17.678687
Traceback (most recent call last):
  File "C:\Program Files\pycharm\PyCharm Community Edition 2021.3.1\plugins\python-ce\helpers\pydev\pydevd.py", line 1483, in _exec
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "C:\Program Files\pycharm\PyCharm Community Edition 2021.3.1\plugins\python-ce\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 358, in <module>
    train(opt)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 284, in train
    grd_x, grd_y = getGrd(img_in)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 37, in getGrd
    grd_x = data_in.clone()
AttributeError: 'numpy.ndarray' object has no attribute 'clone'

Process finished with exit code 1

3.07新进展：传进函数的数据是numpy类型的，而clone收tensor类型的函数，

于是把后面调用的数据改成tensor类型

img_in = torch.from_numpy(img_in)

报错：

self.img_in = torch.from_numpy(np.transpose(img_in, (2, 0, 1)) / 255.0).float()

TypeError: expected np.ndarray (got Tensor)

修改：

为：

 又回到梯度的错误了，但是换了一个问题

 发现图像输入和网络输入维度不一致，用unsqueeze函数和 .squeeze()给图像改维数，新错误，

 哭了。。。心情不太好，决定还是回到前面的问题吧

==>> Mon Mar  7 21:18:55 2022 [Iteration 1 Epoch 450]    Loss = 12893.431641
Traceback (most recent call last):
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 359, in <module>
    train(opt)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 286, in train
    grd_x, grd_y = getGrd(img_in)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 37, in getGrd
    grd_x = grd_x[:, :, 0:-1, :] - data_in[:, :, 1:, :]
IndexError: too many indices for tensor of dimension 3
==>> Mon Mar  7 21:18:56 2022 [Iteration 1 Epoch 500]    Loss = 10786.927734
==>> Mon Mar  7 21:18:57 2022 [Iter 1/8]   loss = 10786.927734
====>> Mon Mar  7 21:18:57 2022   Pass time: 0:00:18.375648

Process finished with exit code 1

在网上搜不到具体方法，反正就是维度的问题吧，采用debug查看了维度，输入图像的维度和

input_net, target_net 的维度不一样，明天继续

 input_net, target_net = Variable(batch[0]), Variable(batch[1], requires_grad=False) 

好吧，是中间batch的维度都不对。