60x Speedup on IUV2FBC Function. Greatly speeds up converting a 2D person image to a 3D model.

[davidleejy]

Referenced in #98
Hello,

This PR gives a 60x speed up to IUV2FBC (in DensePose/detectron/utils/densepose_methods.py). IUV2FBC function converts an IUV point to an XYZ point on a SMPL model. This will convenience users who would like to make a 3D person out of a 2D person image (e.g. in virtual reality).

Timings

Run IUV2FBC on 125 IUV points in 'DensePoseData/demo_data/demo_dp_single_ann.pkl' ...

Original:

8629403 function calls in 5.227 seconds
Ordered by: cumulative time
   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    5.227    5.227 <string>:1(<module>)
        1    0.001    0.001    5.227    5.227 <ipython-input-40-02014517a6cb>:32(f)
      125    0.088    0.001    5.219    0.042 densepose_methods.py:270(IUV2FBC)
    35578    0.430    0.000    5.064    0.000 densepose_methods.py:62(barycentric_coordinates_exists)
   108588    1.486    0.000    4.536    0.000 numeric.py:1591(cross)

Sped up:

39129 function calls in 0.081 seconds
Ordered by: cumulative time
   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    0.081    0.081 <string>:1(<module>)
        1    0.001    0.001    0.081    0.081 <ipython-input-39-6b6ed951a68c>:32(f)
      125    0.014    0.000    0.079    0.001 densepose_methods.py:174(IUV2FBC_fast)
      125    0.035    0.000    0.059    0.000 densepose_methods.py:135(barycentric_coordinates_fast)
      375    0.008    0.000    0.020    0.000 numeric.py:1591(cross)

Run IUV2FBC on 80000 IUV points ...

Original: Estimated ~53 minutes (80000 points / 125 points * 5 secs ~= 53 mins)

Sped up: 1 minute (timing obtained from a profiled run)

PR Contribution

notebooks/DensePose-Fast-IUV-2-XYZ.ipynb : Is a demo. Also addresses issue #87 .

detectron/utils/densepose_methods.py : Contains sped-up codes. Backward-compatible since original functions not modified. New functions: IUV2FBC_fast and barycentric_coordinates_fast.

DensePoseData/demo_data_2/* : Data given for the convenience of running DensePose-Fast-IUV-2-XYZ.ipynb .

[davidleejy]

@nvrv @ralpguler
Loved using Densepose and would be grateful for the opportunity to contribute to this repository. The following sheds more light on how the speed-up came about:

• Main source of speed-up is batching numpy cross product computations (https://stackoverflow.com/questions/1988091/poor-numpy-cross-performance) executed frequented in barycentric_coordinates_exists() and barycentric_coordinates(). Instead of calling numpy.cross() on 2 vectors at a time, I called it on a bunch of vectors in a single call.

• Secondary source of speed-up is removing duplicate computations in barycentric_coordinates_exists() and barycentric_coordinates().

• The result is a single function barycentric_coordinates_fast().

• New code is functionally identical to original code in that it produces the same results.

• End-users call a new function IUV2FBC_fast() that takes in the same parameters as the original function IUV2FBC().
  IUV2FBC_fast( self, I_point , U_point, V_point)
  IUV2FBC( self, I_point , U_point, V_point)

• I admit the demo notebook I gave is kind of messy in hindsight. I could have removed a lot of cells to help better convey this improvement. I could clean it up if it'd help the reviewers & maintainers. :)

Feel free to comment with questions and/or suggestions :)