Any ideas on how to generate depth map faster?

[nvinhphuc]

I have to generate depth-map for a large dataset, so I need to do it faster. According to my observation, the depth map generation is slow because of the z-buffer algorithm run on CPU (costs 3s per image), while the PRNet process the vertices from the image so fast. So I wonder if there are any ways to faster the process (multi-thread, multi-core, etc.)?

[nwdxbx]

@intheroom hi,I also want get the depth map,can you tell me how to get it, thanks!

[nvinhphuc]

@intheroom hi,I also want get the depth map,can you tell me how to get it, thanks!

you can use this command: python demo.py -i <inputDir> -o <outputDir> --isDlib True --isDepth True

[nwdxbx]

@intheroom Thank you ver much. I will try it soon.

[zhuomanliu]

I tried the cython version for generating depth maps, and it only cost 0.0035s per frame.

[Guyume]

@zhuomanliu I have the same problem of generating depth map, but I am new to pyhton, does the cython version have the source code of getting depth map or should write a new function, can you give me some hint?

[baoyu-yuan]

@zhuomanliu hello, I have try the code you referred to generate depth image, but I get different results with the original z-buffer code here. Have you ever encountered this situation ?

[gjd2017]

@zhuomanliu @baorave, I also encounter this situation.