rectangle_perimeter with NaN params causes fatal memory leak

[eyaler]

Description:

Following code raises RAM usage to 100% within a few seconds and hangs the (64GB) machine. I had the NaN's due to an invalid bounding box calculation.

Way to reproduce:

import numpy as np
from skimage.draw import rectangle_perimeter
rectangle_perimeter((np.nan, np.nan), (np.nan, np.nan))

Version information:

3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Windows-10-10.0.22621-SP0 (I am actually on Windows 11)
scikit-image version: 0.21.0
numpy version: 1.24.3

[lagru]

Hmm, I can't reproduce the RAM bombing on my machine. Instead I get

skimage/draw/draw.py:959: RuntimeWarning: invalid value encountered in cast
  top_left = np.round(top_left).astype(int)
skimage/draw/draw.py:960: RuntimeWarning: invalid value encountered in cast
  bottom_right = np.round(bottom_right).astype(int)
skimage/draw/draw.py:271: RuntimeWarning: invalid value encountered in cast
  r = np.round(r).astype(int)
skimage/draw/draw.py:272: RuntimeWarning: invalid value encountered in cast
  c = np.round(c).astype(int)

and this output

(array([-9223372036854775808, -9223372036854775808, -9223372036854775808,
        -9223372036854775808]),
 array([-9223372036854775808, -9223372036854775808, -9223372036854775808,
        -9223372036854775808]))

I am a bit surprised because from a casual look there isn't much that might be platform specific to the involved code. Maybe

scikit-image/skimage/draw/draw.py

Lines 959 to 960 in a45f2da

	top_left = np.round(top_left).astype(int)
	bottom_right = np.round(bottom_right).astype(int)

is behaving different for you? Could you check the output of

np.round((np.nan, np.nan)).astype(int)

on your machine?

In any case, we should probably introduce a guard that checks for NaN, which in this case shouldn't have any noticeable performance implications.

[lagru]

Oh, and thanks for the report @eyaler!

[eyaler]

@lagru

> np.round((np.nan, np.nan)).astype(int)
<ipython-input-3-ea42ef702220>:1: RuntimeWarning: invalid value encountered in cast
  np.round((np.nan, np.nan)).astype(int)
Out[3]: array([-2147483648, -2147483648])

then you have in https://github.com/scikit-image/scikit-image/blob/v0.21.0/skimage/draw/draw.py#L929
top_left -= 1

so let do:
a = np.round((np.nan, np.nan)).astype(int); a -= 1; print(a)
and we get:
array([2147483647, 2147483647])

that's one big rectangle!

[lagru]

Yes, this doesn't seem to happen on my hardware because np.round((np.nan, np.nan)).astype(int) returns array([-9223372036854775808, -9223372036854775808]) for me. That number is negative and get's treated differently.

Are you interested in making a PR that adds a check for NaN's to address this issue? A simple if np.isnan(...): raise ... + a test should suffice. :D

[lagru]

Re my comment above, I actually forgot that I already made a PR for this #7115. 🙈 😄

[github-actions]

Hello scikit-image core devs! There hasn't been any activity on this issue for more than 180 days. I have marked it as "dormant" to make it easy to find.
To our contributors, thank you for your contribution and apologies if this issue fell through the cracks! Hopefully this ping will help bring some fresh attention to the issue. If you need help, you can always reach out on our forum
If you think that this issue is no longer relevant, you may close it, or we may do it at some point (either way, it will be done manually).