4 ms·
I had completely the opposite take-away - but then I have some experience of photography, and sadly, brightly coloured lampshades no longer surprise me. I thin
by TheOtherHobbes 11y ago
I had completely the opposite take-away - but then I have some experience of photography, and sadly, brightly coloured lampshades no longer surprise me.
I think it's a good illustration of the limits of statistical approaches, and why "It's harder than it looks" applies.
This is about as good as it's going to get without genuine object recognition, knowledge of real-world lighting and colour, and awareness of photographic styles.
It might be possible for a system to learn all of the above, but it's going to need a bigger and probably pre-partitioned training set, and a much more complex model.
- logicallee 11y agoI'm curious, for you is the difference obvious? (is the right-hand one obviously reality in each case?)
- eutectic 11y ago[citation needed]
- Houshalter 11y agoTo be useful, it doesn't have to be perfect. It can have humans provide it with reference photos, select colors for different parts, or look at several different versions of the same image and decide which one looks best.