Not really, as noticed in the comments, this is mainly due to an implementation bug with the image downsizing algorithms in the ML libraries:
Look like developers of ML libraries are not always image processing experts 
Coo, what fun! I came across this image:
I sampled it with ImageMagick:
magick c.jpg -sample 267x233 s.jpg
And the result is this:
… which is undeniable proof that Leo de Vinci was a time-traveller. Or that he secretly liked toes. Or something.
The concerned cat turning into a smug dog with a raincoat is too funny.
Most definitely. Too many papers leave me scratching my head.
da Vinci’s time doppelganger no doubt.
Thanks for the correction. I couldn’t remember the painter’s name, and was too lazy to look it up, and made my best guess.
If anyone doesn’t realise: I edited the Mona Lia image replacing one pixel in every 64 with the appropriate pixel from my toes image. IM’s simple and fast “-sample” picks out just those pixels.
I saved the edited Mona Lisa as a lossy JPEG, so that messes up the embedded toes image. When I save it losslessly, “-sample” reproduces the toes exactly.