datum stats and transform with resize failed to run on part of PascalVOC dataset

[shaojun]

hi,

I'm using version 0.2.2, Ubuntu 20.04.3 LTS on x86.

I've downloaded part of officical PASCAL VOC dataset: VOCtrainval_11-May-2012.tar, untar it can see 17125 pictures under voc20212/JPEGImages.

I created a datum project and run datum stats:

datum import -n voc2012 -f voc ../dataset_data/pascalVOC/VOCtrainval_11-May-2012/VOCdevkit/VOC2012/
#Source 'voc2012' with format 'voc' has been added to the project
datum stats

But got below error:

/home/shao/.local/lib/python3.8/site-packages/datumaro/components/operations.py:1031: RuntimeWarning: overflow encountered in uint_scalars
M2 = m_a + m_b + delta ** 2 * count_a * count_b / (count_a + count_b)
/home/shao/.local/lib/python3.8/site-packages/datumaro/components/operations.py:1033: RuntimeWarning: overflow encountered in uint_scalars
count_a + count_b,
/home/shao/.local/lib/python3.8/site-packages/datumaro/components/operations.py:1035: RuntimeWarning: overflow encountered in uint_scalars
M2 / (count_a + count_b - 1)
2022-01-07 15:09:15,977 INFO: Writing project statistics to 'statistics.json'

For keep moving, I tried to resize for uniform all images size:

datum transform -t resize -- -dw 800 -dh 600

it ran a while, then see error:

2022-01-07 15:20:45,303 INFO: Transforming...
Killed

could you check?

[zhiltsov-max]

Hi, I was able to reproduce and fix both problems. The warnings are correct, but they should not really affect the pixel mean and std stats, though. Thank you for reporting the issues!