Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

underflow encountered when using RescaleIntensity

See original GitHub issue

🐛Bug

To reproduce

Download this file, and attempt to rescale between -1, and 1. output.nii.gz

import torchio as tio
import nibabel as nib

# PATH to output.nii.gz
Resection_name = 'output.nii.gz'
Resection_img = tio.ScalarImage(Resection_name)

Rescale = tio.RescaleIntensity( (-1,1) )

Resection_img = Rescale(Resection_img)

Expected behavior

Expected to Resection_img.data to be a torch tensor which has minimum -1 and maximum of 1

Actual behavior

Traceback (most recent call last):
  File "test.py", line 19, in <module>
    Resection_img = Rescale(Resection_img)
  File "/Users/gabriellakamlish/anaconda3/envs/my_Env/lib/python3.8/site-packages/torchio/transforms/transform.py", line 121, in __call__
    transformed = self.apply_transform(subject)
  File "/Users/gabriellakamlish/anaconda3/envs/my_Env/lib/python3.8/site-packages/torchio/transforms/preprocessing/intensity/normalization_transform.py", line 53, in apply_transform
    self.apply_normalization(subject, image_name, mask)
  File "/Users/gabriellakamlish/anaconda3/envs/my_Env/lib/python3.8/site-packages/torchio/transforms/preprocessing/intensity/rescale.py", line 54, in apply_normalization
    image.set_data(self.rescale(image.data, mask, image_name))
  File "/Users/gabriellakamlish/anaconda3/envs/my_Env/lib/python3.8/site-packages/torchio/transforms/preprocessing/intensity/rescale.py", line 77, in rescale
    array /= array_max  # [0, 1]
FloatingPointError: underflow encountered in true_divide

System info

Output of python <(curl -s https://raw.githubusercontent.com/fepegar/torchio/master/print_system.py):

Platform: macOS-10.13.6-x86_64-i386-64bit
TorchIO:  0.18.29
PyTorch:  1.6.0
NumPy:    1.19.4
Python:   3.8.5 (default, Sep  4 2020, 02:22:02) 
[Clang 10.0.0 ]

Issue Analytics

State:
Created 2 years ago
Comments:5 (5 by maintainers)

Top GitHub Comments

1reaction

fepegarcommented, Apr 13, 2021

@romainVala you’re right! It is the same error. Apparently, an underflow happens when a value is too small to be represented by the computer. I suppose the implemented workaround is to use the actual smallest number that can be represented. Anyway, it’s very low-level and doesn’t seem to be a problem for us.

@GabriellaKamlish, this will be solved in v0.18.33. Please upgrade and try again.

@all-contributors please add @GabriellaKamlish for bug

0reactions

allcontributors[bot]commented, Apr 13, 2021

@fepegar

I’ve put up a pull request to add @GabriellaKamlish! 🎉

Top Results From Across the Web

Numpy error: underflow encountered in exp - Stack Overflow

FloatingPointError : underflow encountered in exp. The array I'm trying to define is as follows: time = np.arange(length) window ...

What is the significance of underflow during parameter update ...

I am getting an underflow error when the model is calculating how much to update parameters by multiplying learning rate, momentum, velocity, ...

Issues · fepegar/torchio · GitHub

... Topics Collections Trending Skills GitHub Sponsors Open source guides Connect with others The ReadME Project Events Community forum GitHub Education...

oblique illumination for live cell nanoscale imaging - bioRxiv

ABSTRACT To determine the molecular and/or mechanical basis of cell migration using live. 19 cell imaging tools, it is necessary to ...

UC San Diego - eScholarship.org

B: rescale intensity. ... vessels under flow, stained with MitoTracker Red. ... also has been found to have protective effect towards rat SMC...