The one on the right...
Both columns have two pics each with a caption above and one below that are potentially confusing ....but with the left hand column you have to look through all three pics and all the captions before getting enough context to really know which caption is for which pic.

With the RH side column you know for sure when you look at the first pic that the captions describe the pic above them.

>>460690
this + it's an established pattern, file explorers etc all use the bottom-label pattern so better to stick to that when making potentially ambiguous design
if this is for display, see if you can add some sort of interactivity or animation when they appear that helps segment the hierarchy