8000 Merge pull request #28122 from anntzer/aggunclip · matplotlib/matplotlib@24c3f1a · GitHub
[go: up one dir, main page]
Skip to content

Commit 24c3f1a

Browse files
greglucasgreglucas
authored
Merge pull request #28122 from anntzer/aggunclip
Disable clipping in Agg resamplers.
2 parents ed17e00 + c7eb84a commit 24c3f1a

File tree

8 files changed

+78
-152
lines changed

8 files changed

+78
-152
lines changed

‎extern/agg24-svn/include/agg_span_image_filter_gray.h

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -397,7 +397,9 @@ namespace agg
397397
fg += weight * *fg_ptr;
398398

399399
fg >>= image_filter_shift;
400+
#ifndef MPL_DISABLE_AGG_GRAY_CLIPPING
400401
if(fg > color_type::full_value()) fg = color_type::full_value();
402+
#endif
401403

402404
span->v = (value_type)fg;
403405
span->a = color_type::full_value();
@@ -491,8 +493,10 @@ namespace agg
491493
}
492494

493495
fg = color_type::downshift(fg, image_filter_shift);
496+
#ifndef MPL_DISABLE_AGG_GRAY_CLIPPING
494497
if(fg < 0) fg = 0;
495498
if(fg > color_type::full_value()) fg = color_type::full_value();
499+
#endif
496500
span->v = (value_type)fg;
497501
span->a = color_type::full_value();
498502

@@ -593,8 +597,10 @@ namespace agg
593597
}
594598

595599
fg /= total_weight;
600+
#ifndef MPL_DISABLE_AGG_GRAY_CLIPPING
596601
if(fg < 0) fg = 0;
597602
if(fg > color_type::full_value()) fg = color_type::full_value();
603+
#endif
598604

599605
span->v = (value_type)fg;
600606
span->a = color_type::full_value();
@@ -701,8 +707,10 @@ namespace agg
701707
}
702708

703709
fg /= total_weight;
710+
#ifndef MPL_DISABLE_AGG_GRAY_CLIPPING
704711
if(fg < 0) fg = 0;
705712
if(fg > color_type::full_value()) fg = color_type::full_value();
713+
#endif
706714

707715
span->v = (value_type)fg;
708716
span->a = color_type::full_value();

‎lib/matplotlib/image.py

Lines changed: 9 additions & 92 deletions
Original file line numberDiff line numberDiff line change
@@ -457,93 +457,21 @@ def _make_image(self, A, in_bbox, out_bbox, clip_bbox, magnification=1.0,
457457
# input data is not going to match the size on the screen so we
458458
# have to resample to the correct number of pixels
459459

460-
# TODO slice input array first
461-
a_min = A.min()
462-
a_max = A.max()
463-
if a_min is np.ma.masked: # All masked; values don't matter.
464-
a_min, a_max = np.int32(0), np.int32(1)
465460
if A.dtype.kind == 'f': # Float dtype: scale to same dtype.
466-
scaled_dtype = np.dtype(
467-
np.float64 if A.dtype.itemsize > 4 else np.float32)
461+
scaled_dtype = np.dtype("f8" if A.dtype.itemsize > 4 else "f4")
468462
if scaled_dtype.itemsize < A.dtype.itemsize:
469463
_api.warn_external(f"Casting input data from {A.dtype}"
470464
f" to {scaled_dtype} for imshow.")
471465
else: # Int dtype, likely.
466+
# TODO slice input array first
472467
# Scale to appropriately sized float: use float32 if the
473468
# dynamic range is small, to limit the memory footprint.
474-
da = a_max.astype(np.float64) - a_min.astype(np.float64)
475-
scaled_dtype = np.float64 if da > 1e8 else np.float32
476-
477-
# Scale the input data to [.1, .9]. The Agg interpolators clip
478-
# to [0, 1] internally, and we use a smaller input scale to
479-
# identify the interpolated points that need to be flagged as
480-
# over/under. This may introduce numeric instabilities in very
481-
# broadly scaled data.
482-
483-
# Always copy, and don't allow array subtypes.
484-
A_scaled = np.array(A, dtype=scaled_dtype)
485-
# Clip scaled data around norm if necessary. This is necessary
486-
# for big numbers at the edge of float64's ability to represent
487-
# changes. Applying a norm first would be good, but ruins the
488-
# interpolation of over numbers.
489-
self.norm.autoscale_None(A)
490-
dv = np.float64(self.norm.vmax) - np.float64(self.norm.vmin)
491-
vmid = np.float64(self.norm.vmin) + dv / 2
492-
fact = 1e7 if scaled_dtype == np.float64 else 1e4
493-
newmin = vmid - dv * fact
494-
if newmin < a_min:
495-
newmin = None
496-
else:
497-
a_min = np.float64(newmin)
498-
newmax = vmid + dv * fact
499-
if newmax > a_max:
500-
newmax = None
501-
else:
502-
a_max = np.float64(newmax)
503-
if newmax is not None or newmin is not None:
504-
np.clip(A_scaled, newmin, newmax, out=A_scaled)
505-
506-
# Rescale the raw data to [offset, 1-offset] so that the
507-
# resampling code will run cleanly. Using dyadic numbers here
508-
# could reduce the error, but would not fully eliminate it and
509-
# breaks a number of tests (due to the slightly different
510-
# error bouncing some pixels across a boundary in the (very
511-
# quantized) colormapping step).
512-
offset = .1
513-
frac = .8
514-
# Run vmin/vmax through the same rescaling as the raw data;
515-
# otherwise, data values close or equal to the boundaries can
516-
# end up on the wrong side due to floating point error.
517-
vmin, vmax = self.norm.vmin, self.norm.vmax
518-
if vmin is np.ma.masked:
519-
vmin, vmax = a_min, a_max
520-
vrange = np.array([vmin, vmax], dtype=scaled_dtype)
521-
522-
A_scaled -= a_min
523-
vrange -= a_min
524-
# .item() handles a_min/a_max being ndarray subclasses.
525-
a_min = a_min.astype(scaled_dtype).item()
526-
a_max = a_max.astype(scaled_dtype).item()
527-
528-
if a_min != a_max:
529-
A_scaled /= ((a_max - a_min) / frac)
530-
vrange /= ((a_max - a_min) / frac)
531-
A_scaled += offset
532-
vrange += offset
469+
da = A.max().astype("f8") - A.min().astype("f8")
470+
scaled_dtype = "f8" if da > 1e8 else "f4"
471+
533472
# resample the input data to the correct resolution and shape
534-
A_resampled = _resample(self, A_scaled, out_shape, t)
535-
del A_scaled # Make sure we don't use A_scaled anymore!
536-
# Un-scale the resampled data to approximately the original
537-
# range. Things that interpolated to outside the original range
538-
# will still be outside, but possibly clipped in the case of
539-
# higher order interpolation + drastically changing data.
540-
A_resampled -= offset
541-
vrange -= offset
542-
if a_min != a_max:
543-
A_resampled *= ((a_max - a_min) / frac)
544-
vrange *= ((a_max - a_min) / frac)
545-
A_resampled += a_min
546-
vrange += a_min
473+
A_resampled = _resample(self, A.astype(scaled_dtype), out_shape, t)
474+
547475
# if using NoNorm, cast back to the original datatype
548476
if isinstance(self.norm, mcolors.NoNorm):
549477
A_resampled = A_resampled.astype(A.dtype)
@@ -564,21 +492,10 @@ def _make_image(self, A, in_bbox, out_bbox, clip_bbox, magnification=1.0,
564492
# Apply the pixel-by-pixe F543 l alpha values if present
565493
alpha = self.get_alpha()
566494
if alpha is not None and np.ndim(alpha) > 0:
567-
out_alpha *= _resample(self, alpha, out_shape,
568-
t, resample=True)
495+
out_alpha *= _resample(self, alpha, out_shape, t, resample=True)
569496
# mask and run through the norm
570497
resampled_masked = np.ma.masked_array(A_resampled, out_mask)
571-
# we have re-set the vmin/vmax to account for small errors
572-
# that may have moved input values in/out of range
573-
s_vmin, s_vmax = vrange
574-
if isinstance(self.norm, mcolors.LogNorm) and s_vmin <= 0:
575-
# Don't give 0 or negative values to LogNorm
576-
s_vmin = np.finfo(scaled_dtype).eps
577-
# Block the norm from sending an update signal during the
578-
# temporary vmin/vmax change
579-
with self.norm.callbacks.blocked(), \
580-
cbook._setattr_cm(self.norm, vmin=s_vmin, vmax=s_vmax):
581-
output = self.norm(resampled_masked)
498+
output = self.norm(resampled_masked)
582499
else:
583500
if A.ndim == 2: # interpolation_stage = 'rgba'
584501
self.norm.autoscale_None(A)

‎lib/matplotlib/tests/baseline_images/test_image/imshow_masked_interpolation.png

-1 Bytes
Loading

0 commit comments

Comments
 (0)
0