OSV_q The size of tensor a (704) must match the size of tensor b (320) at non-singleton dime

Claire_Shang

已于 2022-03-17 10:04:23 修改

阅读量1.6k

点赞数 2

文章标签：深度学习 pytorch 人工智能

于 2022-03-16 12:20:04 首次发布

本文链接：https://blog.csdn.net/Claire_wanqing/article/details/123523310

版权

图像维度 batch_size loss 模型训练参数调整

关键词由CSDN通过智能技术生成

====>> Wed Mar 16 12:16:26 2022   Pass time: 0:02:06.581927
==>>   save samples 13!

barbara
Load Data: barbara
====>>   Build Net

====>>   Load I model

====>>   Train
torch.Size([1, 3, 544, 704]) torch.Size([1, 3, 544, 704])
Traceback (most recent call last):
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 410, in <module>
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 301, in train
    lc_time = time.asctime(time.localtime(time.time()))
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 97, in loss_func
    u11 = (f1 - dive1 + b2).to(torch.float32)
RuntimeError: The size of tensor a (704) must match the size of tensor b (320) at non-singleton dimension 3

Process finished with exit code 1

应该是图像维度不对,发现问题：原来定义的这个代码固定了图像维度

q1 = torch.zeros([1, 3, 320, 320])

于是改为输入图像的维度，成功

 q1 = torch.zeros(f.shape)
 q1 = q1.to(device)
 b1 = b2 = q2 = q1

已经开始运行啦，进入调参数过程

不知道为什么，当我增加batch_size=5后（原来是1），就报错

====>>   Train
Traceback (most recent call last):
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 395, in <module>
    train(opt)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 295, in train
    print(outstr + '    Loss = %f' % loss.data.cpu().numpy())
UnboundLocalError: local variable 'loss' referenced before assignment

Process finished with exit code 1

http://t.csdn.cn/VwT8M

分析局部变量与全局变量，在循环外加入这行语句

 loss = None

但是报错了

====>>   Train
Traceback (most recent call last):
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 395, in <module>
    train(opt)
  File "C:/Users/shang/Desktop/STDN_LI/TVGnet/OSV_q.py", line 295, in train
    print(outstr + '    Loss = %f' % loss.data.cpu().numpy())
AttributeError: 'NoneType' object has no attribute 'data'

Process finished with exit code 1

http://t.csdn.cn/EwpTS

也不行，先不改batch_size吧，先调别的参数试试