refactor code to replace np.clip

This commit is contained in:
KIU Shueng Chuan 2021-03-20 08:09:29 +08:00
parent 12aee3e47d
commit 8d74dfe11b

View File

@ -998,6 +998,27 @@ def solveBilinearTransform(points1, points2):
return matrix
def clip_array(arr, vmin, vmax, out=None):
# replacement for np.clip due to regression in
# performance since numpy 1.17
# https://github.com/numpy/numpy/issues/14281
if out is None:
out = np.empty_like(arr)
if vmin is not None:
arr = np.core.umath.maximum(arr, vmin, out=out)
if vmax is not None:
arr = np.core.umath.minimum(arr, vmax, out=out)
# np.core.umath.clip performs slightly better than
# the above on platforms compiled with GCC (e.g. Linux),
# but worse for CLANG (e.g. macOS) and MSVC (Windows)
return out
def rescaleData(data, scale, offset, dtype=None, clip=None):
"""Return data rescaled and optionally cast to a new dtype.
@ -1010,16 +1031,12 @@ def rescaleData(data, scale, offset, dtype=None, clip=None):
else:
dtype = np.dtype(dtype)
vmin, vmax = None, None
if dtype.kind in 'ui':
lim = np.iinfo(dtype)
if clip is None:
# don't let rescale cause integer overflow
vmin, vmax = lim.min, lim.max
else:
vmin, vmax = max(clip[0], lim.min), min(clip[1], lim.max)
elif clip is not None:
vmin, vmax = clip
clip = lim.min, lim.max
clip = max(clip[0], lim.min), min(clip[1], lim.max)
if np.can_cast(data, np.float32):
work_dtype = np.float32
@ -1030,11 +1047,8 @@ def rescaleData(data, scale, offset, dtype=None, clip=None):
d2 *= scale
# Clip before converting dtype to avoid overflow
# regression in np.clip performance since numpy 1.17
if vmin is not None:
np.core.umath.maximum(d2, vmin, out=d2)
if vmax is not None:
np.core.umath.minimum(d2, vmax, out=d2)
if clip is not None:
clip_array(d2, clip[0], clip[1], out=d2)
# don't copy if no change in dtype
data = d2.astype(dtype, copy=False)