imshow alpha interpolation looks wrong

Hi, not much of an issue, but it seems that imshow is not interpolating RGBA images in the proper way. There was some discussion about a similar issue in #9906 which ended up on a documentation note:

imshow expects RGB images adopting the straight (unassociated) alpha representation.

If that's the case imshow should still switch to associated colors (premultiply alpha) before interpolation to avoid artefacts like the one below, shouldn't it?

import matplotlib.pyplot as plt
import numpy as np

fig, (axl, axr) = plt.subplots(1, 2)

# full green image
img = np.zeros((5, 5, 4))
img[...,1] = np.ones((5, 5))
# transparent above main diagonal
img[...,3] = np.tril(np.ones((5, 5), dtype=np.uint8))

axl.imshow(img, interpolation="none")
axr.imshow(img, interpolation="bilinear")

plt.show()

it seems something Agg specific, it doesn't happen e.g. with the cairo backend

It looks like it is compositing it into black (0, 0, 0) rather than white (1, 1, 1)....

@tacaswell

@tacaswell maybe my description can be a bit misleading. Black fringing is a common artifact when some part of a graphic pipeline uses premultiplied alpha while others don't, but I'm not entirely sure that's what's happening here.

matplotlib/lib/matplotlib/image.py

Lines 480 to 496 in 0334c04

    
           else: 
        
               # Always convert to RGBA, even if only RGB input 
        
               if A.shape[2] == 3: 
        
                   A = _rgb_to_rgba(A) 
        
               elif A.shape[2] != 4: 
        
                   raise ValueError("Invalid dimensions, got %s" % (A.shape,)) 
        
               output = np.zeros((out_height, out_width, 4), dtype=A.dtype) 
        
               alpha = self.get_alpha() 
        
               if alpha is None: 
        
                   alpha = 1.0 
        
               _image.resample( 
        
                   A, output, t, _interpd_[self.get_interpolation()], 
        
                   self.get_resample(), alpha, 
        
                   self.get_filternorm(), self.get_filterrad())

That is where we re-sample RBGA images for rendering. From looking at that I suspect that this is an Agg bug, but we should be hitting this code path in cairo as well...

I suspect the fix here is to change to output starting as (1, 1, 1, 0), which will make it look right more often (as the default patch color is white), but should dig into it a bit more. We may want to re-sample the RGB and A seperatly like we do with the color-mapped data.

Crude fix for issue matplotlib#11316

@tacaswell

@tacaswell was right about doing the alpha and rgb interpolation separatly.

While building a test for this issue, I came across a problem with the interpolation keyword when savefig() is used with the SVG format. See issue #12065.

fix for issue matplotlib#11316

@dabana

This was fixed by #12062 / #13559. Thanks @dabana !

Navigation Menu

Repository navigation

imshow alpha interpolation looks wrong #11316

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	else:
	# Always convert to RGBA, even if only RGB input
	if A.shape[2] == 3:
	A = _rgb_to_rgba(A)
	elif A.shape[2] != 4:
	raise ValueError("Invalid dimensions, got %s" % (A.shape,))

	output = np.zeros((out_height, out_width, 4), dtype=A.dtype)

	alpha = self.get_alpha()
	if alpha is None:
	alpha = 1.0

	_image.resample(
	A, output, t, _interpd_[self.get_interpolation()],
	self.get_resample(), alpha,
	self.get_filternorm(), self.get_filterrad())

imshow alpha interpolation looks wrong #11316

Description

Activity

fargiolas commented on May 26, 2018

tacaswell commented on May 29, 2018

fargiolas commented on May 29, 2018

tacaswell commented on May 29, 2018

dabana commented on Sep 7, 2018

dabana commented on Sep 7, 2018

tacaswell commented on May 17, 2019

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions