Depth

[Michiel VO]

Hi all,

I am Michiel, a master student and for my thesis I am working with light field camera's (Lytro Illum) for depth perception and distance calculation. Currently I am stuck trying to generate a usable depth map

Would anyone happen to have experience generatign a depth map from decode lytro illum images?

Thank you

[Ian Faust]

I recommend looking through Vincent Qin's repository (https://github.com/Vincentqyw/Depth-Estimation-Light-Field). He collected some of the codes submitted from the CVPR competition in 2017. Though spinning up the results from Wanner and Goldluecke 2014 is also pretty straightforward (Christopher Hanhe's python implementation is well-coded https://github.com/hahnec/depthy).

Most if not all the algorithms whose papers I've read can be classified in 2-3 parts.

1) Localized depth finding

2) Depth information propagation (MRF, SGM, TV-L smoothing, etc.)

3?) Occlusion correction (sometimes)

or they just train a special deep neural net (of which training sets seem to be very limited, making it very likely they are overfitted).

If you are wanting to decode lpf files directly you can also probably find that too.

Hope this helps.

-Ian