[bug] trying to save a tensor to image, changes its content #1719
Description
I have a float tensor in range 0-1 that I want to save as binary image.
I get this stack trace:
b=batch(4)
C:\Program Files\JetBrains\PyCharm Community Edition 2019.1.3\helpers\pydev\pydevconsole.py:75: RuntimeWarning: invalid value encountered in double_scalars
m = save_main_module('<input>', 'pydevconsole')
c = b[2]
c.max()
Out[136]: tensor(1., dtype=torch.float64)
c = c.view(-1 ,256, 256)
c.max()
Out[138]: tensor(1., dtype=torch.float64)
torchvision.utils.save_image(c, r"C:\shrec\temp.png")
Traceback (most recent call last):
File "C:\Users\t-maserr\AppData\Local\Programs\Python\Python37\lib\site-packages\PIL\Image.py", line 2533, in fromarray
mode, rawmode = _fromarray_typemap[typekey]
KeyError: ((1, 1, 32), '|u1')
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "C:\Users\t-maserr\AppData\Local\Programs\Python\Python37\lib\site-packages\IPython\core\interactiveshell.py", line 3319, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "<ipython-input-139-fe16f2ca20a4>", line 1, in <module>
torchvision.utils.save_image(c, r"C:\shrec\temp.png")
File "C:\Users\t-maserr\AppData\Local\Programs\Python\Python37\lib\site-packages\torchvision\utils.py", line 104, in save_image
im = Image.fromarray(ndarr)
File "C:\Users\t-maserr\AppData\Local\Programs\Python\Python37\lib\site-packages\PIL\Image.py", line 2535, in fromarray
raise TypeError("Cannot handle this data type")
TypeError: Cannot handle this data type
c.max()
Out[140]: tensor(255., dtype=torch.float64)
Notice how the maximum value of the tensor is 1 before trying to save it, then it changes to 255.
It somehow scaled from 0:1 to the 127:255 range.
Latest PyTorch + TorchVision, on Python 3.7, Windows.