Gouraud shading with tripcolor renders incorrectly as SVG in Jupyter notebook

[CorbinFoucart]

Bug report

Bug summary

When the matplotlib backend is set to 'SVG' in a Jupyter Notebook, it seems that matplotlib does not render 'gouraud' shading correctly.

Code for reproduction

I've uploaded a minimal example of my problem here:
https://nbviewer.jupyter.org/github/CorbinFoucart/FEMexperiment/blob/master/app/mpl_bug.ipynb

Actual outcome

Note the 'gouraud' shading with SVG in the notebook. If I save the output to file and view the output, it is similarly not shaded, which indicates to me that this is not a Jupyter issue, nor a browser issue.

Matplotlib version

• Operating system: OSX El Capitan 10.11.6
• Matplotlib version: 2.2.2
• Matplotlib backend (print(matplotlib.get_backend())): module://ipykernel.pylab.backend_inline
• Python version: 3.6.4
• Jupyter version:

ipykernel==4.8.2
jupyter==1.0.0
jupyter-client==5.2.3
jupyter-console==5.2.0
jupyter-core==4.4.0

I installed matplotlib in a virtual environment with pip.

[WeatherGod]

SVG is a vector based format. Does anybody know if SVG supports gouraud shading in the format specs? If not, then what you have to do is specify "rasterized=True" in the arguments for tripcolor() sothat the tripcolor is not saved as a vector but as a bitmap within the SVG file.

…

On Tue, May 29, 2018 at 1:05 AM, Avinash Sharma ***@***.***> wrote: The problem is not specific to "SVG" backend or to Jupyter. Even on saving the plot in "SVG" format with a different backend, the saved "SVG" has the same problem. Try this: import numpy as np import matplotlib as mpl import matplotlib.pyplot as plt import matplotlib.tri as mtri #plt.switch_backend('svg') plt.switch_backend('nbAgg') # Mesh N = 10 x, y = np.linspace(0, 1, N), np.linspace(0, 1, N) X, Y = np.meshgrid(x, y) # simple function over the mesh Z = X**2 + Y**2 # triangulate tri = mtri.Triangulation(X.ravel(), Y.ravel()) # plot fig, ax = plt.subplots(1, 2) # default ax = fig.add_subplot(1, 2, 1) ax.tripcolor(tri, Z.ravel(), cmap=plt.cm.coolwarm) ax.set_title('default') ax.axis('equal') # gouraud ax = fig.add_subplot(1, 2, 2) ax.tripcolor(tri, Z.ravel(), cmap=plt.cm.coolwarm, shading='gouraud') ax.set_title('gouraud') ax.axis('equal') plt.savefig('output.svg', dpi= 1000) plt.savefig('output.png', dpi= 1000) plt.show() Although the *.show()* output and *png* file have correct shading but the *svg* file is still incorrect. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11321 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-FIOwr0E33c9HslbHMd-wzfCMu84ks5t3NcqgaJpZM4UParc> .

[tacaswell]

It has the right methods,

matplotlib/lib/matplotlib/backends/backend_svg.py

Lines 664 to 773 in 1966a7a

	def draw_gouraud_triangle(self, gc, points, colors, trans):
	# This uses a method described here:
	#
	# http://www.svgopen.org/2005/papers/Converting3DFaceToSVG/index.html
	#
	# that uses three overlapping linear gradients to simulate a
	# Gouraud triangle. Each gradient goes from fully opaque in
	# one corner to fully transparent along the opposite edge.
	# The line between the stop points is perpendicular to the
	# opposite edge. Underlying these three gradients is a solid
	# triangle whose color is the average of all three points.
	writer = self.writer
	if not self._has_gouraud:
	self._has_gouraud = True
	writer.start(
	'filter',
	id='colorAdd')
	writer.element(
	'feComposite',
	attrib={'in': 'SourceGraphic'},
	in2='BackgroundImage',
	operator='arithmetic',
	k2="1", k3="1")
	writer.end('filter')
	avg_color = np.sum(colors[:, :], axis=0) / 3.0
	# Just skip fully-transparent triangles
	if avg_color[-1] == 0.0:
	return
	trans_and_flip = self._make_flip_transform(trans)
	tpoints = trans_and_flip.transform(points)
	writer.start('defs')
	for i in range(3):
	x1, y1 = tpoints[i]
	x2, y2 = tpoints[(i + 1) % 3]
	x3, y3 = tpoints[(i + 2) % 3]
	c = colors[i][:]
	if x2 == x3:
	xb = x2
	yb = y1
	elif y2 == y3:
	xb = x1
	yb = y2
	else:
	m1 = (y2 - y3) / (x2 - x3)
	b1 = y2 - (m1 * x2)
	m2 = -(1.0 / m1)
	b2 = y1 - (m2 * x1)
	xb = (-b1 + b2) / (m1 - m2)
	yb = m2 * xb + b2
	writer.start(
	'linearGradient',
	id="GR%x_%d" % (self._n_gradients, i),
	x1=short_float_fmt(x1), y1=short_float_fmt(y1),
	x2=short_float_fmt(xb), y2=short_float_fmt(yb))
	writer.element(
	'stop',
	offset='0',
	style=generate_css({'stop-color': rgb2hex(c),
	'stop-opacity': short_float_fmt(c[-1])}))
	writer.element(
	'stop',
	offset='1',
	style=generate_css({'stop-color': rgb2hex(c),
	'stop-opacity': "0"}))
	writer.end('linearGradient')
	writer.element(
	'polygon',
	id='GT%x' % self._n_gradients,
	points=" ".join([short_float_fmt(x)
	for x in (x1, y1, x2, y2, x3, y3)]))
	writer.end('defs')
	avg_color = np.sum(colors[:, :], axis=0) / 3.0
	href = '#GT%x' % self._n_gradients
	writer.element(
	'use',
	attrib={'xlink:href': href,
	'fill': rgb2hex(avg_color),
	'fill-opacity': short_float_fmt(avg_color[-1])})
	for i in range(3):
	writer.element(
	'use',
	attrib={'xlink:href': href,
	'fill': 'url(#GR%x_%d)' % (self._n_gradients, i),
	'fill-opacity': '1',
	'filter': 'url(#colorAdd)'})
	self._n_gradients += 1
	def draw_gouraud_triangles(self, gc, triangles_array, colors_array,
	transform):
	attrib = {}
	clipid = self._get_clip(gc)
	if clipid is not None:
	attrib['clip-path'] = 'url(#%s)' % clipid
	self.writer.start('g', attrib=attrib)
	transform = transform.frozen()
	for tri, col in zip(triangles_array, colors_array):
	self.draw_gouraud_triangle(gc, tri, col, transform)
	self.writer.end('g')

but claims to not support it

matplotlib/lib/matplotlib/backends/backend_svg.py

Line 264 in 1966a7a

self._has_gouraud = False

There probably needs to be some git-forensics to sort the history out here.

[avinashmnit30]

The problem is with the SVG backend. Its giving correct output with cario backend. Here's the mplcairo implementation to generate SVG file and it seems to work.

import numpy as np
import matplotlib as mpl
mpl.use("module://mplcairo.qt")
import matplotlib.pyplot as plt
import matplotlib.tri as mtri 
from cairocffi import *
#%matplotlib notebook
#plt.switch_backend('mplcairo')
print(mpl.get_backend())
%config InlineBackend.figure_format = 'svg'
# Mesh
N = 10
x, y = np.linspace(0, 1, N), np.linspace(0, 1, N)
X, Y = np.meshgrid(x, y)

# simple function over the mesh
Z = X**2 + Y**2

# triangulate
tri = mtri.Triangulation(X.ravel(), Y.ravel())
plt.Figure()
# plot
fig, ax = plt.subplots(1, 2)


# default
ax = fig.add_subplot(1, 2, 1)
ax.tripcolor(tri, Z.ravel(), 
                   cmap=plt.cm.coolwarm)
ax.set_title('default')
ax.axis('equal')


# gouraud
ax = fig.add_subplot(1, 2, 2)
ax.tripcolor(tri, Z.ravel(), 
                   cmap=plt.cm.coolwarm, 
                   shading='gouraud')
ax.set_title('gouraud')
ax.axis('equal')
plt.savefig('outputmplqt.svg')
#plt.savefig('outputmplqt.png')
plt.show()

[tacaswell]

@avinashmnit30 Is a diff of the two svg files understandable?

[avinashmnit30]

@tacaswell "Cario backend" is using PNG image within <image> element of the SVG. So, it's not actually generating Gouraud-triangles in SVG.
Also, the "SVG backend" implementation of Gouraud triangles is incorrect. Its mentioned that it is following http://www.svgopen.org/2005/papers/Converting3DFaceToSVG/index.html paper. But the implementation is incorrect.

Example:

Code for reproduction: https://matplotlib.org/2.0.1/examples/pylab_examples/quadmesh_demo.html

SVG Backend Output:

Using correct implementation:

[jklymak]

Is this still an open issue given #11378?

[timhoffm]

This is the output on current master.

The non-masked version is now correct. However, there is a difference for triangles that are adjacent to the masked area: Our SVG implementation does not add transparency to these triangles. Probably it should, but I'm not an expert in these things.

[anntzer]

fwiw cairo (and thus mplcairo) renders this "correctly" as SVG, but does so by rasterizing the entire mesh.

[cover-me]

The current version of svg_backend doesn't work. I think rasterizing the mesh is the only way to achieve Gouraud shading in SVG files at this moment. Because even if triangles get right Gouraud colors, the anti-aliasing issue (#5694) between two adjacent patches would out make the output useless.

Something like this:

[github-actions]

This issue has been marked "inactive" because it has been 365 days since the last comment. If this issue is still present in recent Matplotlib releases, or the feature request is still wanted, please leave a comment and this label will be removed. If there are no updates in another 30 days, this issue will be automatically closed, but you are free to re-open or create a new issue if needed. We value issue reports, and this procedure is meant to help us resurface and prioritize issues that have not been addressed yet, not make them disappear. Thanks for your help!

[anntzer]

I think this is still valid.

[katya-zossi]

I recently had an issue with svg format and 'gouraud' shading.
Thanks for the tip to add 'rasterized=True' to the tripcolor() function.