It's clear from the example you highlighted that there just isn't much contrast for a computer algorithm to recognise anything as an edge (even though we as humans can discern it - that's because we're actually 'intelligent', not just 'artificially' so).
Absolutely, @Wedge009 - this all the way
That R&D on edge detection algorithms sounds fascinating! Yeah, I believe the big quantum leap in deep learning image handling in the last decade or so was the development of GANs (Generative Adversarial Networks) - that's when this stuff really started to take off. For anyone interested, there's some further reading here