Python API¶
This section includes information for using the pure Python API of
bob.ip.optflow.liu
.
Root module functions¶
ConjugateGradient (CG) based Implementation¶
Ce Liu’s Optical Flow implementations using CG

bob.ip.optflow.liu.cg.
flow
(i1, i2, [alpha=0.02, [ratio=0.75, [min_width=30, [n_outer_fp_iterations=20, [n_inner_fp_iterations=1, [n_cg_iterations=50]]]]]]) > (u, v, w2)¶ This method computes the dense optical flow field using a coarsetofine approach. C++ code running under this call is extracted from the old version (pre Aug 1, 2011) of Ce Liu’s homepage and should give the exact same output as the Matlab equivalent.
Note
This variant does not use the Successive OverRelaxation (SOR) that was implemented on August 1st., 2011 by C. Liu, but the old version based on ConjugateGradient (CG).
Parameters:
 i1
 First input frame (grayscale/double image)
 i2
 Second input frame (same dimension and type of the first frame)
 alpha
 [optional] Regularization weight
 ratio
 [optional] Downsample ratio
 min_width
 [optional] Width of the coarsest level
 n_outer_fp_iterations
 [optional] The number of outer fixed point iterations
 n_inner_fp_iterations
 [optional] The number of inner fixed point iterations
 n_cg_iterations
 [optional] The number of conjugategradient (CG) iterations
Returns a tuple containing three 2D double arrays with the same dimensions as the input images:
 u
 Output velocities in
x
(horizontal axis).  v
 Output velocities in
y
(vertical axis).  warped_i2
 i2 as estimated by the optical flow field from i1
SuccessiveOverRelaxation (SOR) based Implementation¶
Ce Liu’s Optical Flow implementations using SOR

bob.ip.optflow.liu.sor.
flow
(i1, i2, [alpha=1.0, [ratio=0.5, [min_width=40, [n_outer_fp_iterations=4, [n_inner_fp_iterations=1, [n_sor_iterations=20]]]]]]) > (u, v, w2)¶ This method computes the dense optical flow field using a coarsetofine approach. C++ code running under this call is extracted from the old version (pre Aug 1, 2011) of Ce Liu’s homepage and should give the exact same output as the Matlab equivalent.
Note
This variant uses the Successive OverRelaxation (SOR) that was implemented on August 1st., 2011 by C. Liu.
Parameters:
 i1
 First input frame (grayscale/double image)
 i2
 Second input frame (same dimension and type of the first frame)
 alpha
 [optional] Regularization weight
 ratio
 [optional] Downsample ratio
 min_width
 [optional] Width of the coarsest level
 n_outer_fp_iterations
 [optional] The number of outer fixed point iterations
 n_inner_fp_iterations
 [optional] The number of inner fixed point iterations
 n_sor_iterations
 [optional] The number of successiveoverrelaxation (SOR) iterations
Returns a tuple containing three 2D double arrays with the same dimensions as the input images:
 u
 Output velocities in
x
(horizontal axis).  v
 Output velocities in
y
(vertical axis).  warped_i2
 i2 as estimated by the optical flow field from i1